We use data from the Chinese Health and Retirement Longitudinal Study (CHARLS) 2011–2018, a nationally representative survey of adults aged 45 and older in China. CHARLS provides detailed retrospective information on fertility history, educational attainment and urban–rural residence for respondents born across several decades. This study focuses on participants with at least one living child, who comprised 96.4% of the original sample. After excluding 924 respondents without living children from the initial 25,586 participants, our final sample consisted of 24,662 individuals. On average, these respondents were 56.74 years of age and had 2.5 living children.

Although CHARLS is a longitudinal survey, we pool data from all available waves (2011, 2013, 2015 and 2018) to maximise coverage and minimise missing information. Fertility histories, which include information on age at first birth and number of living children, were collected retrospectively and are time-invariant once reported. Therefore, pooling observations does not duplicate individuals’ fertility histories, but it does allow us to include respondents who entered in later waves or whose demographic information (e.g., education or urban–rural residence) was updated in subsequent interviews.

The sample covers multiple birth cohorts who experienced different socioeconomic and policy contexts during their reproductive years: pre-1940 (N = 2678), 1940–1950 (N = 4864), 1950–1960 (N = 7286) and post-1960 (N = 8291). The pre-1940 cohort reached reproductive ages during the 1950s-1960s, largely before the introduction of systematic fertility policy interventions, and thus experienced relatively unrestricted fertility under traditional demographic regimes. The 1940–1950 cohort experienced their peak reproductive years (1960s–1980s) during the initial implementation of the “later, longer, fewer” campaigns of the 1970s that promoted delayed marriage, longer birth intervals and smaller families. The 1950–1960 cohort lived through their reproductive years (1970s–1990s) during the implementation and early years of China’s one-child policy beginning in 1979. The post-1960 cohort experienced their reproductive years (1980s–2000s) during the policy window characterised by established enforcement mechanisms and differentiated urban–rural implementation. This cohort variation allows us to examine how fertility timing patterns varied across different policy contexts, and to assess whether different socioeconomic groups exhibited similar or divergent demographic patterns as policy environments evolved throughout China’s fertility transition.

Researchers have employed various approaches to measuring first birth timing and delayed childbearing, each of which has distinct advantages and limitations in different research contexts. The most prevalent approach involves making simple temporal comparisons of mean ages at first birth across cohorts or time periods (Beaujouan et al., 2023; Beck et al., 2024; Kohler et al., 2002), which effectively captures broad trends of fertility postponement, but does not distinguish timing variations within birth cohorts. Meanwhile, many studies treat age at first birth as a continuous variable without establishing explicit definitions of “delayed” fertility (Giuntella et al., 2022; Goisis and Sigle-Rushton, 2014; Tropf and Mandemakers, 2017), limiting their ability to categorise individuals or examine timing effects systematically. Some researchers have used fixed thresholds, such as distinguishing between younger and older mothers using the 27th birthday as a cutoff (Frejka and Sardon, 2006), though this approach may not account for changing fertility norms. In contexts where fertility patterns shift across cohorts, relative measures that adjust for cohort-specific norms may provide more meaningful distinctions between early and delayed childbearing, allowing researchers to identify individuals who postpone fertility relative to their contemporaries, rather than relying on a single pre-determined age cut point.

We adopt a relative approach that accommodates changing fertility norms across China’s rapid demographic transition. We operationalise delayed first birth in two complementary ways: standard delayed first birth (first birth exceeding the cohort-specific mean age) and very delayed first birth (first birth exceeding the cohort-specific mean plus one standard deviation). This measurement approach acknowledges the significant cohort changes in fertility norms throughout China’s transition, with a “delayed” birth in the 1940s differing substantially from one in the 1980s, while using cohort-specific thresholds accounts for period-specific social and economic conditions influencing typical childbearing patterns.

Family size is measured using the self-reported number of living children for each respondent, which captures completed fertility for our sample of adults aged 45 and older. Educational attainment is categorised into three levels: low education (no formal education to elementary school), middle education (middle school only) and higher education (high school, vocational school and college or above).

Urban–rural residence is coded as a binary indicator based on respondents’ current place of residence and administrative hukou status as recorded in the survey. This measure reflects respondents’ residential and registration status at the time of the interview, rather than during their childbearing years. Although this approach is standard in studies using CHARLS, it may not fully capture the historical residence of individuals who migrated after completing fertility. However, because most cohorts in our analysis (born before 1960) completed childbearing prior to the large-scale urban–rural migration that accelerated in the 1990s (Chan and Zhang, 1999; Wu and Treiman, 2004), potential misclassification bias is likely minimal.

To directly connect our analysis of delayed first birth patterns to the implications for eldercare, we have developed metrics that potentially capture different dimensions of intergenerational support structures. These metrics translate fertility timing differences into practical measures of eldercare availability and implications of potential strain on family support systems.

Our primary measure, Child’s Age When Parent Reaches 80, is calculated as: Child’s     Age     When     Parent     Reaches     80 = 80 − Mother / Father’s     Age     at     First     Birth (1) This metric represents the age of the eldest child when the mother reaches age 80, an important threshold when intensive care needs often emerge. Lower values indicate that the children will be younger and potentially more capable of providing intensive care when their parents reach advanced ages.

Building on this measure, we have developed three additional metrics to capture different aspects of intergenerational support dynamics: Potential Caregiving Years (measuring pre-retirement years available for caregiving), Care Overlap Index (measuring potentially problematic competing demands) and Care Timing Quality Index (assessing optimal timing for caregiving capacity). Detailed formulas for these supplementary metrics are included in Appendix A. We compare these metrics across educational levels (low, middle and high) and urban–rural residence to assess how socioeconomic status shapes not just family size, but also the timing of eldercare needs relative to the adult children’s life course stages.

We employ Kitagawa decomposition to quantify how much of the family size differences between socioeconomic groups (educational levels or urban–rural status) are attributable to differences in first birth timing versus differences in family size within fertility timing groups. For each cohort, the decomposition partitions the family size difference between two groups (e.g., higher vs lower education) into two components. The composition effect captures the portion attributable to differences in the distribution of first birth timing differences between groups, indicating how much of the family size gap would change if both groups had the same timing-specific family sizes but differed in the timing of first births. The rate effect captures the portion attributable to differences in completed family sizes within fertility timing groups, indicating how much of the gap would remain if both groups had the same first birth timing distribution but differed in family size conditional on timing. This approach allows us to isolate the contribution of first birth timing differences to overall socioeconomic disparities in family size.

Formally, for educational differences, the decomposition is: Difference = ∑ i [ ( r 1 i − r 2 i ) ( c 1 i + c 2 i ) / 2 ] + ∑ i ( c 1 i − c 2 i ) ( r 1 i + r 2 i ) / 2 ] (2) where r represents rates (family sizes), c represents compositions (proportions of delayed/non-delayed fertility) and i represents categories (delayed vs non-delayed fertility). The first two terms represent the rate effect, and the latter represents the composition effect. The same approach is applied to decompose urban–rural differences.

We perform this decomposition separately for each birth cohort to examine how the contribution of first birth timing differences to socioeconomic differences in family size has evolved across different periods. Additionally, we compare results using both standard and very delayed fertility measures to distinguish between moderate and extreme fertility postponement. All analyses are conducted using Stata 16SE. Survey weights are used.

Socioeconomic gradients in first birth timing existed throughout China’s fertility transition, though they emerged differently for educational than for urban–rural stratification. Table 1 shows a persistent and strong educational gradient in first birth timing. Among the pre-1940s cohort, those with higher education were substantially more likely to delay their first birth, with a gap of 39.3 percentage points between the lowest and highest education groups (40.7% vs 80%). This gap widened to 46.7 percentage points for the 1940–1950 cohort (39% vs 85.7%) before stabilising at around 50.6 percentage points for more recent cohorts. This gradient persisted throughout China’s fertility transition, with higher education being consistently and strongly associated with delayed childbearing across all cohorts.

The contrast is particularly evident for very delayed first births (first births more than one standard deviation above the mean age). Among the post-1960 cohort, 39.2% of adults with the highest educational level, compared to just 10.3% of those with the lowest educational level, experienced a very delayed first birth – a gap that has increased by approximately 60% since the pre-1940 cohort.

Unlike the educational gradient, which was strong even in the earliest cohort, urban–rural differences in first birth timing emerged gradually over cohorts. Table 1 reveals that among the pre-1940s cohort, there were minimal urban–rural differences in the share with a delayed first birth (46.9% vs 43.5%, a gap of only 3.4 percentage points). However, this gap widened to 15.0 percentage points for the 1950–1960s cohort (60.7% vs 45.7%) and remained substantial, at 14.3 percentage points, for the post-1960s cohort (58.7% vs 44.4%). This growing urban–rural divide reflects the increasingly differential experience of the demographic transition in China’s urban versus rural areas, which is a product of uneven policy implementation, expanding urban educational and employment opportunities and rapidly changing urban norms regarding family formation.

Despite differences in fertility timing patterns, all population segments experienced substantial fertility declines and convergence towards smaller family sizes across cohorts. Table 2 documents this universal decline, with the mean family size decreasing from 4.12 children among the pre-1940 cohort to 2.18 children among the post-1960 cohort (Panel A). This decline occurred across all socioeconomic groups. Panel B shows that even among the highest fertility group (those with no formal education), the average family size declined from 4.35 to 2.68 children, while among the lowest fertility group (college educated), it declined from 3.12 to 1.68 children. Similarly, Panel C shows that both urban and rural populations experienced substantial reductions in family size, with the average family size of urban families decreasing from 3.78 to 2.01 children and that of rural families declining from 4.38 to 2.42 children across cohorts. This convergence towards smaller family sizes occurred despite persistent socioeconomic differences in absolute levels. The educational gap between the highest and lowest educated groups remained relatively stable (ranging from −1.00 to −1.31 children), while the urban–rural gap narrowed from −0.60 to −0.41 children.

This convergence towards similar family sizes despite divergent timing patterns raises a critical question: How much of the remaining socioeconomic differences in family size can be attributed to differences in first birth timing? We address this question through decomposition analysis. Table 3 shows how much of the family size differences between socioeconomic groups and regions can be attributed to differences in first birth timing.

Despite this strong educational gradient in first birth timing, differences in age at first birth explain only 3.4% to 9.6% of the education-based family size gap across cohorts using our standard measure, with very similar results (3.4% to 9.6%) being generated when using the more restrictive measure of very delayed first births (Panel B, Table 3). The consistency across the two timing measures suggests that differences in first birth timing explain only a small portion (less than 10%) of educational differences in completed fertility. The remaining portion (more than 90%) of the educational gap operates through mechanisms other than first birth timing, which could include differences in birth spacing or stopping behaviour, or differential responses to policy constraints. However, our analysis cannot distinguish among these possibilities. The decomposition shows that the vast majority of educational differences in family size stem from the “rate effect”, meaning that highly educated individuals have fewer children than less educated individuals regardless of whether they delay entry to parenthood. Among those who delay the first birth, individuals with higher education still have substantially fewer children than lower educated individuals; similarly, among those who do not delay the first birth, highly educated individuals have smaller family sizes than their less educated counterparts.

Importantly, these educational differences have persisted despite China’s fertility policies, which suggests that even within policy constraints, different educational groups display distinct fertility behaviours. While China’s fertility policies created upper limits on family size, enforcement varied by time period, geography and social group, and compliance was not uniform across educational levels (Gu et al., 2007; Niu and Qi, 2020; Zhang and Zhao, 2023; Zhao and Zhang, 2018). China’s policies may have limited how much timing differences could translate into family size differences, but they did not eliminate educational variation entirely. This contrasts with contexts characterised by voluntary fertility decline, where delayed childbearing typically reduces completed fertility through biological or social constraints. The results indicate that educational differences in family size cannot be primarily attributed to differences in first birth timing. While we cannot directly observe the underlying mechanisms, this pattern suggests that factors beyond the timing of entry into parenthood explain the majority of educational fertility differentials.

Delayed first births contribute substantially to explaining urban–rural family size differences throughout China’s fertility transition, although the magnitude of the contribution varies depending on the timing measure used. Using our standard measure of delayed first birth, differences in first birth timing explain 6.6% of the urban–rural family size gap for the pre-1940 cohort, 22.3% of this gap for the 1940–1950 cohort during an intense period of policy implementation and 13.0% of this gap for the post-1960 cohort (Panel A, Table 3). However, when using the more restrictive measure of very delayed first births, the contributions are generally smaller (ranging from 4.9% to 7.5% across cohorts), suggesting that moderate postponement plays a more important role than extreme postponement in urban–rural fertility differences. This finding quantifies how the urban fertility decline has incorporated meaningful first birth postponement as one mechanism for achieving smaller families in urban areas, whereas this pattern is largely absent in rural areas.

Table 4 presents the results of our decomposition analysis for the post-1960 cohort, showing how much of the socioeconomic differences in caregiving metrics can be attributed to differences in first birth timing.

For the urban–rural divide, Table 4 shows that differences in first birth timing account for a substantial share of disparities in intergenerational age structures. Specifically, 42.3% of the urban–rural gap in a child’s age when the parent reaches age 80 is attributable to the higher prevalence of delayed first birth among urban residents. In other words, nearly half of the urban advantage in having younger children during the parents’ advanced ages stems from later entry into parenthood. We note, however, that this analysis focuses on the age of the first (eldest) child, whereas rural families with multiple children may experience different eldercare dynamics through having more potential caregivers, though this distinction operates through family size rather than through timing mechanisms. For the post-1960 cohort, whose average number of living children has already declined to around 2.2, such quantity differences are smaller, making timing a more salient dimension of inequality. Similarly, differences in first birth timing explain 44.8% of the urban–rural difference in potential caregiving years and 38.2% of the difference in the Care Overlap Index.

For educational differences, first birth timing explains over half of the disparities in eldercare timing. The total educational difference in the child’s age when the parent reaches age 80 is 2.54 years, with differences in first birth timing explaining 56.7% of this gap (1.44 years), meaning that the children of highly educated parents are nearly 1.5 years younger when their parents reach advanced ages solely due to delayed childbearing patterns. In terms of potential caregiving years, the total educational advantage is 2.54 years, with first birth timing explaining 58.3% of this difference (1.48 years), meaning that children of educated parents have nearly 1.5 additional years of pre-retirement caregiving capacity. Similarly, first birth timing accounts for 52.1% of the educational difference in the Care Overlap Index, explaining 1.28 years of the total 2.45-year difference in intensive care timing.

Overall, the evidence presented in Table 4 suggests that first birth timing is likely a major mechanism through which socioeconomic status shapes eldercare contexts. While we previously found that first birth timing explains a relatively modest portion of socioeconomic differences in family size (less than 10% for education and up to 22% for urban–rural status), it accounts for 40–60% of the socioeconomic differences in eldercare metrics. This contrast highlights how fertility timing has become a pathway through which socioeconomic factors influence intergenerational support structures in later life.

To assess the robustness of our findings, we conducted additional analyses using different definitions of first birth timing. First, we compared our standard definition (births exceeding the cohort-specific mean age) with a more stringent “very delayed first birth” measure (births exceeding the mean plus one standard deviation), as shown in Table 2. While the magnitude of timing’s contribution varied – notably for urban–rural differences, where it decreased from 13.0% to 7.5% for the post-1960 cohort – the fundamental pattern remained consistent. First birth timing explained a modest but meaningful portion of family size differences, with direct fertility quantum effects (the rate effect) consistently dominating.

We extended our sensitivity analysis by implementing alternative thresholds including the 75th and 90th percentiles of the cohort-specific age distributions (results not shown). These percentile-based approaches confirmed our main findings while showing slightly smaller composition effects, with first birth timing explaining only 4–7% of urban–rural family size differences when using the 90th percentile threshold (compared to 7–12% when using the mean-based definition). Importantly, across all measurement approaches, the rate effect remained the dominant mechanism, accounting for 88–96% of socioeconomic differences in family size.

For our eldercare metrics, we tested whether the observed socioeconomic patterns were sensitive to the definition of delayed first birth. When using the median as the threshold, the contribution of first birth timing to eldercare disparities remained substantial, explaining 40–54% of educational differences and 32–38% of urban–rural differences in the timing of children’s caregiving availability. The 75th percentile threshold yielded similar results (38–52% for educational differences and 35–40% for urban–rural differences). Even with fixed age cutoffs (age 27 for the post-1960 cohort), which are less sensitive to cohort-specific fertility patterns, first birth timing still explained 36–48% of educational differences and 30–35% of urban–rural differences in eldercare metrics.

Despite these contributions, our study has several limitations. First, our analysis cannot fully disentangle the effects of state fertility policies from voluntary socioeconomic influences on first birth timing. Although we observe distinct timing patterns across cohorts exposed to different policy regimes, the precise causal mechanisms driving these patterns remain difficult to isolate. Future research could address this limitation by linking fertility histories with local-level policy implementation data or by exploiting regional policy variation to better distinguish policy effects from socioeconomic or cultural factors.

Second, our measure of completed fertility is based on the number of living children reported at the time of the interview and excludes respondents without surviving offspring. This approach inevitably conditions fertility outcomes on child survival. Because child mortality declined substantially after the 1950s (Feeney and Yuan, 1994; Zeng et al., 2001; Zhao et al., 2014), this likely results in a modest underestimation of completed fertility for the oldest cohort, who experienced reproduction before the mortality decline. However, socioeconomic and urban–rural differences in child survival were relatively modest for later cohorts (Cai, 2012; Zeng et al., 2002; Zhao et al., 2014), suggesting that survival bias is unlikely to drive major socioeconomic patterns in these cohorts. Nonetheless, older, less educated and rural respondents were somewhat more likely to experience child loss, which may lead us to slightly understate their completed fertility and overstate convergence across groups. Accordingly, our estimates should be interpreted as reflecting gradients in surviving rather than total children. Future research could reconstruct full parity histories or integrate historical mortality data to yield more accurate estimates of completed fertility across cohorts.

Third, our conceptualisation of fertility timing focuses exclusively on the age at first birth and does not capture other key dimensions of fertility tempo. Birth spacing, or the interval between successive births, represents an equally important component that our analysis cannot address. This limitation is particularly relevant in the Chinese context, where the later, longer, fewer campaign of the 1970s explicitly regulated birth intervals, and where variation in spacing could have affected completed fertility independently of first birth timing. We also do not examine the age at last birth or the overall duration of the childbearing period, both of which shape fertility outcomes and intergenerational structures. Rural families permitted to have two or more children may have exhibited different spacing and stopping patterns than urban families, offering additional pathways through which socioeconomic factors may have influenced family size. The finding that first birth timing explains only a modest portion of socioeconomic fertility differences suggests that these unmeasured aspects of fertility tempo, together with preferences for family size, likely play important roles. Future research incorporating complete fertility schedules, including birth intervals and stopping behaviour, would provide a more comprehensive understanding of how tempo effects operate across socioeconomic groups in China’s policy-constrained setting.

Fourth, our eldercare metrics assume that both parents and their children survive to age 80, an assumption that may not hold equally across socioeconomic groups. Higher educated and urban residents typically have longer life expectancies, making them more likely to reach advanced ages at which care needs intensify. Their children also tend to experience better survival and health outcomes, potentially amplifying observed disparities in caregiving availability. Future research should account for differential mortality patterns to produce more accurate projections of eldercare capacity across socioeconomic and regional contexts.

Finally, while we find systematic differences in intergenerational age structures, we cannot directly observe how these differences translate into actual caregiving behaviour. Our metrics capture potential care availability but not realised care provision, which depends on factors such as migration, financial resources, gender roles and cultural expectations. Longitudinal studies following families over time could help to clarify how demographic potential translates into observed caregiving arrangements and the extent to which socioeconomic and regional inequalities in family structure manifest as unequal care in later life.

Despite these limitations, our findings point to the need for ageing and family policies that explicitly account for the demographic timing inequalities embedded in China’s fertility transition. Differences in first birth timing across socioeconomic groups have produced uneven intergenerational age structures, meaning that parents’ care needs and children’s caregiving capacities no longer align uniformly across the population. Policies expanding community-based eldercare and long-term care insurance should therefore prioritise rural and lower educated adults, who are most likely to experience caregiving deficits when their children are already near or past retirement age. At the same time, the rise of delayed fertility among urban and educated populations implies that caregiving will increasingly coincide with midlife labour force participation. As China gradually raises the statutory retirement age, reforms that allow partial retirement, flexible caregiving leave and pension credits for family care could help reduce this life course tension. Because caregiving responsibilities remain heavily gendered, with women bearing the brunt of both eldercare and childcare duties, integrating childcare and eldercare policy – e.g., through workplace flexibility, care leave benefits and community-level service hubs – would support intergenerational care continuity and mitigate the compounding disadvantages faced by midlife women. Together, these directions suggest that effective responses to population ageing must move beyond fertility promotion or the provision of old age support in isolation to incorporate a coordinated life course approach that recognises how the timing of family formation shapes caregiving needs decades later.

The Potential Caregiving Years metric is calculated as: Potential  Caregiving  Years=60−Child’s  Age  When  Parent  Reaches  80+5where 60 represents the typical retirement age in China, Child’s Age When Parent Reaches 80 = 80 - Mother’s Age at First Birth, The addition of five years accounts for the period of parental care needs beginning at approximately age 75.

This metric estimates the number of pre-retirement years available for the eldest child to provide care during the parent’s period of highest need (approximately ages 75–80). Higher values indicate more potential caregiving capacity before the child faces their own retirement transition. For example, a value of 10 indicates that the child will have approximately 10 years between the onset of their parent’s intensive care needs (around age 75) and their own retirement (age 60), potentially allowing for more consistent caregiving support.

The Care Overlap Index is calculated as: Care  Overlap  Index=max(0, Child’s  Age  When  Parent  Reaches  80−55+10)where 55 represents the early retirement/late career threshold in China, 10 represents approximately 10 years of parental high care needs (from age 75–85) and max(0, x) ensures the index cannot be negative.

This index measures the number of years when the parent requires intensive care while the child is simultaneously approaching or is in retirement, and is thus potentially facing their own health limitations or competing responsibilities. The formula accounts for parental care needs beginning around age 75 and continuing until approximately age 85, while considering that children above age 55 may begin to face their own age-related constraints on their caregiving capacity. Higher values indicate more problematic timing with greater competing demands. For example, a value of 15 indicates that for approximately 15 years, the parent will require care while the child is in the late career/retirement transition phase, potentially placing significant strain on the caregiving relationship.

The Care Timing Quality Index is calculated as: Care  Timing  Quality  Index  =100−|50−Child’s  Age  When  Parent  Reaches  80|where 50 represents the theoretically optimal age for caregiving capacity, and |x| represents the absolute value function.

This index produces a score from 0–100, with 100 representing optimal timing (the child is age 50 when the parent is age 80). The formula is based on research suggesting that middle-aged adult children (around age 50) may be at an optimal caregiving life stage, as they have likely established their career and potentially completed their childrearing responsibilities, but are not yet facing significant health limitations themselves. The index decreases linearly as the child’s age deviates in either direction from this optimal age. Higher values indicate more favourable care timing. For example, a value of 95 indicates that the child will be close to the optimal caregiving age when the parent reaches advanced age, while a value of 70 indicates a substantial deviation from optimal timing.