Later First Births in Two Generations Push Grandmotherhood Back 13 Years

A woman becomes a grandmother through two first births: her own, then her daughter’s. When first births move later in both generations, the delays accumulate. Dora Costa’s central estimate puts the median age at first grandmotherhood at 43.6 under 1960 first-birth schedules and 56.9 under 2024 schedules—a 13-year increase. The period calculation combines the two generations’ first-birth schedules and, in the version Costa presents, follows one child. She describes that one-child calculation as a lower-bound estimate: with several children, a grandmother is more likely to become one sooner.

The change has accelerated. About seven of the 13 additional years in the median occurred since 2000. The daughter’s timing accounts for much of the current delay: applying the 2024 schedule to daughters alone moves the median by 6.5 years when mothers remain on the 1960 schedule, and by five years when mothers remain on the 1980 schedule. Between 2007 and 2024, the median age at first birth rose 3.2 years; the implied median age at first grandmotherhood rose 5.4 years. In that specific 2007–2024 implied-CDF comparison, the share projected to be grandmothers by age 60 falls from 83% to 67%.

Costa calls this compound delay, but the term needs qualification. In discussion, one participant asked where the compounding occurs in the economic model of a daughter postponing childbirth. Costa distinguished that model from the two-generation arithmetic: “It’s not the compounding per se, it’s the fact that the bite hits at the second delay.” The model concerns the cost to a mother when her daughter delays; compound delay describes how first births in both generations push grandmotherhood later. The distinction matters because the model’s question—whether a mother and daughter would bargain over birth timing—is not the same as the calculation of how age at first grandmotherhood changes when both generations’ schedules move.

The estimates also depend on how the generations’ timing is related. Costa said the initial convolution ignores correlation between a mother’s and daughter’s ages at first birth, then noted that the data suggest the relationship is often positive. If mothers who gave birth later also have daughters who give birth later, ignoring that relationship may understate the delay. The period calculations apply a year’s rates as if a hypothetical cohort faced them, much like a period life-expectancy calculation. The one-child assumption leaves out the possibility that another child could become a parent earlier. The realized-cohort evidence, which tracks actual people, tells a related but distinct story.

Costa’s headline comparison estimates that healthy years shared with a grandchild fell from about 28 under 1960 birth schedules to about 15 under 2024 schedules. One version holds the healthy-life endpoint fixed at age 72.5; the presentation also shows estimates using group-specific health windows and measures that allow the window to move as disability-free life expectancy changes.

Those adjustments reduce, but do not erase, the estimated loss. For women born in the 1930s, the healthy-year gap between college and non-college women was 8.5 years among White women and 9.5 among Black women under a common endpoint. Using group-specific health windows narrows the gaps to seven and 7.9 years. Costa’s measures concern limitations and need for assistance; they do not capture every kind of activity or capacity for grandparenting.

Longer life has not kept pace with delayed first births. By 2024, the increase in the median age at first grandmotherhood exceeded the gain in life expectancy at 65 by 8.3 years. Since 2000, life expectancy at 65 rose 1.8 years, while the median age at first grandmotherhood rose 6.7. Costa said healthy life expectancy has increased too, but not enough to restore the earlier overlap in the calculations shown. Her calculation of what would be needed to recover the 1960 overlap puts the required healthy endpoint at about age 86. The disability-free life-expectancy figures shown for 1970 and 2010 were 74.2 and 76.5.

The loss also varies with the grandchild’s age. In the presentation’s conclusion, Costa reports that 17% of healthy years with a grandchild under 10 are lost, compared with 60% of healthy years with a grandchild aged 14 to 24. A separate care-weighted calculation for ages 14 to 24 reports a 46% loss of healthy overlap and says that this period carried 36% of the grandchild care in the calculation. These figures use different measures and should not be read as the same comparison. Costa suggested that later years may matter for passing on family norms, skills, and support; the study does not measure whether losing that overlap changes children’s development or what they learn from grandparents.

Participants noted that children’s needs change with age: toddlers require physical attention, while older grandchildren may need rides or other kinds of help. Costa treated these possibilities as reasons the later years could matter, not as effects demonstrated by the estimates.

That distinction matters for welfare. A participant questioned whether presenting fewer shared years as a penalty implied that grandmothers would prefer to spend that time with grandchildren. Costa said the study makes no welfare claim: some grandmothers may strongly want that interaction; others may value alternatives more. The estimates describe when overlap changes, not whether that change is good or bad.

The delay is concentrated among college-educated women. Costa’s synthetic-cohort analysis combines Census and American Community Survey data to estimate grandmotherhood by education and race. By age 55, about 30% of college-educated women in both the White and Black groups had become grandmothers, compared with about 60% of non-college women in both groups. In the Health and Retirement Study, the education gap in grandmotherhood by 55 was roughly 40 percentage points and narrowed only to 37 across the cohorts shown.

The two datasets answer different questions. The Census and ACS allow Costa and her coauthor to construct schedules by education, race, and cohort across a long span, but the analysis is synthetic: it infers first-birth timing from women’s ages and the ages of their eldest co-resident children rather than following a mother and daughter through their lives. Some portions of the schedules require extrapolation, which the researchers flag. The HRS records realized grandmotherhood and can be used to study wealth, transfers, and other outcomes, but covers fewer people and cohorts.

Family size compounds the risk of delay. Each child is another chance for a first grandchild; with one child, that child alone determines whether and when grandmotherhood occurs. In Costa’s estimates, moving from two children to one lowers the probability of grandmotherhood by 55 by 11 to 15 percentage points. Moving from three children to two lowers it by four to five points. Among college women, the two-to-one change also pushes first-quartile grandmotherhood 2.6 to 3.1 years later. Fewer children therefore mean not only fewer potential grandchildren, but a greater chance that grandmotherhood arrives late or not at all.

Timing can persist across generations. The researchers’ estimates indicate that when a college-educated mother’s daughter follows the college group’s later schedule, the mother’s probability of being a grandmother by 55 falls by 10 to 15 percentage points relative to the comparison schedule shown. Costa presented this as a result of the synthetic-cohort analysis, not as a directly observed change for every mother-daughter pair.

In the realized-cohort data, the share of women who were grandmothers by 55 declined from 72% among those born in 1940–44 to 53% among those born in 1960–64. By age 70, however, the decline was only three points. For the cohorts observed in the HRS, Costa therefore characterized the shift mainly as later grandmotherhood, rather than a large decline in eventual incidence. She cautioned that younger women who have not yet reached the relevant ages may change that picture.

The HRS also lets the researchers compare siblings born to the same mother. Holding the mother fixed absorbs shared characteristics such as education, race, family size, resources, and health. Among children of college-educated mothers, siblings born ten years apart differed substantially: the younger-born sibling was 10 to 20 percentage points less likely to be a parent by age 30. Costa said the pattern appears for sons as well as daughters and reflects cohort differences, not simply differences between families.

The education results require care in interpretation. A participant questioned why the education gap had remained so large despite changes in women’s opportunities. Costa agreed that the HRS comparison cannot isolate who selects into college or how changing selection alters the meaning of education across cohorts. As college attendance expands, the marginal college graduate is not necessarily comparable to the earlier cohorts. The data establish a persistent gap by education category; they do not by themselves identify its cause.

Costa’s reproductive-control analysis traces a cross-generational effect. A woman born in 1929 may never have had access to the pill, but her daughters did. Earlier access to the pill and abortion affected young women’s early childbearing, schooling, and careers, particularly among college-educated women. If a daughter’s first birth occurs later, her mother becomes a grandmother later too.

To estimate this relationship, Costa and her coauthor use state of birth and birth cohort to measure whether unmarried women aged 18 to 20 could legally consent to the pill or abortion. The outcome is having a first birth by about age 25, among women observed at ages 24 to 26 in the 1960–1980 censuses. The analysis separates White college, White non-college, and Black non-college women; the Black college sample was too small for the comparison. Costa described the legal measure as an intention-to-treat measure: it captures access under the laws in a woman’s birth state, not whether she actually used a method.

The legal-access measure is associated with a 5.0- to 5.4-percentage-point reduction in the probability of a first birth by about 25 among White college women across the specifications shown. The estimates are close whether the researchers add state trends, restrict to non-migrants, or use an alternative legal coding. They find essentially no effect for White non-college women in this combined measure. For Black non-college women, the estimated coefficient is positive under each measure, but Costa cautioned against interpreting it as a clear effect: other programs and limited variation make the group difficult to analyze.

The researchers’ separate estimates suggest that pill access mattered most for college women, while abortion access was more closely associated with early births among non-college women. Costa emphasized a limitation: only 4% of the variation in abortion-consent laws survives the fixed effects in their model. The comparison is therefore suggestive, not a clean way to distinguish the effects of the pill from those of abortion.

In their accounting, reproductive control explains 56% of the decline in early grandmotherhood among White college women between the 1929 and 1949 birth cohorts. About 59% of the estimated effect for the 1949 cohort runs through daughters. The result supports the paper’s cross-generational mechanism: a policy that changes one generation’s timing can alter when the previous generation becomes a grandmother, even if the older woman was never directly exposed to it. It does not mean that reproductive control accounts for all changes in grandmotherhood, or that the same estimate applies to every group.

On days when women report contact with a grandchild, recorded childcare rises from 12 to 54 minutes—a gain of about 42 minutes. Market work falls by 31 minutes, television by 22, and other leisure by nine. The American Time Use Survey identifies grandchild contact on a given day but does not establish whether a woman has grandchildren overall. Costa presented these estimates as descriptive.

The measure has a limit: the survey records primary activities, while childcare can be entered as a secondary activity. A participant noted that a grandparent may care for a child while watching television, so the apparent fall in television may not represent all viewing that coincides with care. Costa agreed that secondary activities are not fully captured and said the appendix examines secondary care without changing the basic result.

The pattern varies with work status. Among employed grandmothers, market work falls by about 39 minutes on contact days, while television does not. Among retired grandmothers, television falls by 43 minutes out of an average 226 minutes. Costa said HRS event-study estimates do not show an abrupt retirement change at a first grandchild’s birth. Retirement is often already underway, and having fewer expected grandchildren can explain at most a quarter of the rise in work shown in their analysis. A participant suggested that the women who retire may also be the women who have grandchildren; Costa agreed that selection is possible.

Care also shifts over the life course. Women born in the 1950s with college education were less likely than non-college women to provide 100 or more hours of grandchild care at ages 55–74, and more likely to provide that level of care to a parent. Costa described this as a shift in the kind of care provided at different ages, not simply an absence of care among college-educated women.

The first grandchild’s arrival is associated with a 28-percentage-point increase in the probability of providing 100 or more hours of grandchild care, while reported money transfers to the new parent rise by $202 a year. The researchers also find a 4.5-percentage-point increase in living within ten miles, much of it involving a mother moving in with her daughter. Costa had noted that the HRS does not allow the researchers to observe long-term changes in geographic distance across a broad set of cohorts.

Care follows the female line more strongly than money does: the reported daughter premium in care is 9.2 percentage points, while there is no comparable daughter premium in money transfers. Participants questioned whether small reported cash transfers captured gifts in kind or were underreported. Costa said the HRS figures reflect what respondents reported and acknowledged that cash transfers could be underreported. The measures do not include every practical contribution a grandparent might make, such as buying clothes or diapers.

Among women aged 65–79, estimated net worth per grandchild rose 55% for college graduates, from $267,000 in 1995–2006 to $413,000 in 2008–2022, in 2022 dollars. For women with less than high school, a high-school credential, or some college, the measure fell by 5% to 13%. Costa said the rise among college graduates reflects higher wealth, not simply smaller families; the calculation includes household wealth, including housing, and therefore reflects both spouses’ resources.

Women without grandchildren are included as zeros. Among college-educated women, the share of wealth held by grandmothers fell from 92% to 81%, and roughly one-fifth of their wealth was held by women with no grandchild. Costa noted that among women aged 65–69 who had no grandchild, three in five were still alive without one at 79. The researchers’ accounting attributes about two-thirds of the education gap in wealth per grandchild to wealth growth rather than fertility. That is a decomposition of the observed change, not a measure of what any individual grandchild receives.

The economic framework separates two reasons a grandmother might value grandchildren. “Existence value” is satisfaction that the family line continues; it rises when the first grandchild arrives but, on its own, predicts no change in time use. “Interaction value” comes from time spent with grandchildren during the grandmother’s remaining healthy years. It predicts that time will be reallocated from work or leisure.

The framework also identifies a conflict in timing. A daughter may benefit from postponing her first birth while building a career; her payoff eventually falls as she gets older. Her mother’s payoff falls with delay because a later-arriving grandchild leaves fewer healthy years for interaction. The daughter’s timing therefore imposes a cost on her mother. Costa calls the resulting issue the “Rotten Adult Kid Problem,” while stressing that the empirical work focuses on what can be tested with the available data.

Why not bargain over an earlier birth in exchange for care or money? Costa and participants discussed several obstacles. A daughter cannot credibly promise a birth date. A mother cannot guarantee future care because her health may fail, and a promise of money or a bequest may be difficult to condition on timing without affecting the grandchildren. Even when a family could make such an arrangement, Costa said, a payment for having a child might be considered repugnant. One participant offered an example of a mother who had promised to provide care in exchange for a daughter having children; Costa allowed that such a bargain can work, but argued that care is the central and difficult-to-commit margin.

The model’s outside options push in opposing directions. A mother with attractive alternatives to caregiving may be less harmed by delay, flattening the timing cost. But if she later provides care without compensation, the value of her forgone work and leisure makes that provision costly. Participants challenged how the first effect should work, including whether a mother with a demanding career would still prefer delay. Costa’s explanation was that better alternatives can make her more accepting of delay, reducing the timing wedge, even as the cost of care itself remains.

The discussion also raised factors the simple model does not resolve. Participants asked how altruism, reciprocity, or a sense of duty to parents might shape family bargaining. Costa said she could not recall data in the study that measured preferences toward children in the continuous way the analysis would need. She argued that changes in economic incentives could help explain changes over time; a participant replied that filial duties might still affect how fertility would look in their absence.

The calibration makes the distinction concrete. The estimated correction to the daughter’s birth timing is the mother’s loss divided by the curvature of the daughter’s payoff from delay. Costa compares an estimated maternal loss of roughly $227 to $1,089 for each year of delay with a much larger annual career return to delaying for college graduates, estimated at $2,900 to $10,000 per year until the early thirties. Under a complete contract, the model moves the first birth only about five weeks earlier—between one week and four and a half months across the assumptions shown. Costa values those weeks to the mother at about $46 per lineage, against a present value of $4,300 to $13,400 in money and care supplied.

Across the discount rates, hours rules, and return assumptions shown, the correction remains under half a year. Costa’s conclusion is that the family disagreement is more likely to be over who provides care after a grandchild arrives than over the date of the first birth.