An evidence synthesis
A synthesis of research findings and primary sources behind the research presented in Birthgap: the peer-reviewed paper in Nature Portfolio, the data behind the research, and the implications for researchers and public understanding.
Read the paper On a Microdemographic Framework for Decomposing Contemporary Fertility DynamicsThe Birthgap documentary follows Stephen J. Shaw on a nine-year research project to identify common structural patterns across low-fertility nations. Given it is atypical for research to be filmed as it uncovers notable discoveries, this synthesis is intended to align the exploration seen in the film with its later peer-reviewed paper. The underlying analysis, drawn from birth records covering more than 300 million mothers across 33 nations, was published in Scientific Reports in 2025 following three rounds of peer review. All data came from national statistical agencies, sourced directly or via the Human Fertility Database.
This page sets out the published findings and the sources behind that research. For each principal finding, you can examine the claim, the data behind it, and where it was peer-reviewed.
The Birthgap Project was started in 2016 by Stephen J. Shaw, a data scientist, to investigate common structural patterns in fertility across low-fertility nations. This approach differs from most demographic research, which primarily examines associations with socioeconomic factors. The documentary follows this investigation and the discoveries made along the way. It is research-based and non-advocacy: it does not recommend or advocate any policies. The film follows Shaw's data exploration and interviews filmed across 24 nations with men and women who want children, who do not want children, and who had once hoped to have children. The core analytical findings were later peer-reviewed and published in Scientific Reports (2025), drawing on a separate, even larger dataset of official birth records covering 33 nations. The core statistical analysis relied on official birth records rather than survey responses.
To the author's knowledge, this research is the first to document, using consistent measurement across national vital-registration data, that family size among mothers who do become mothers is remarkably stable across highly diverse national contexts (Finding 3), and that the corresponding declines in total fertility are therefore driven almost entirely by declining period motherhood rates, rather than mothers having fewer children (Finding 1). No prior formal identification of this pattern, across this range of countries and using this measurement approach, is known to the author. This priority claim is stated openly, and is open to challenge on the basis of prior work the author may not be aware of.
All figures and tables reproduced from Shaw (2025) are used under CC BY-NC-ND 4.0 from Nature Portfolio.
The paper decomposes the Total Fertility Rate as TFR = TMR × CPM, where TMR (Total Maternal Rate) is an indicator of motherhood likelihood and CPM is family size among mothers. Statistical testing (breakpoint co-occurrence, Fisher's Exact Test p=0.548 exact-year, p=0.144 within one year; wavelet coherence, no country statistically significant) found no evidence that TMR and CPM move together; they behave largely independently. Wavelet coherence (Fig. 3, reproduced below) tests whether two time series rise and fall together at matching frequencies over time; here it finds no such relationship between TMR and CPM in any country studied. Mutual-information analysis across 33 economies (1,470 country-years) found only weak dependence (MI ≈ 0.185 bits).
Because TFR blends the two into one number, it obscures which one is actually moving: Monte Carlo modelling found that reporting TMR and CPM separately, instead of TFR alone, raised explanatory power (Adjusted R²) by a mean of +0.385 across 726,000 simulations (p<0.001), and, in the paper's entropy analysis, TFR retained only 51.1% of the information present in its two components (a 48.9% loss, 95% CI 48.86–50.68%).
Why it matters
Because motherhood incidence and family size vary largely independently of one another, any single number that blends them, including TFR, cannot reveal which one is actually changing. Research and policy conclusions drawn from TFR alone risk conflating two distinct trends with different causes.
Source: Shaw (2025), Scientific Reports, Results section, Fig. 3 (reproduced below).
The paper introduces the Microdemographic Framework (MDF), formalising three period-based measures: Total Maternal Rate (TMR), an indicator of the share of women who would become mothers under current fertility patterns; Children per Mother (CPM), average family size among those who do; and Total Childlessness Rate (TCR), the complement of TMR. TFR = TMR × CPM.
The framework's practical value is not limited to the aggregate result reported above (see Core Contribution): decomposing TFR into TMR and CPM improved explanatory power (Adjusted R²) in every one of the 33 nations studied, with country-level mean gains ranging from roughly +0.17 to +0.32 (reproduced below). The improvement is consistent and positive across every country tested, not an artefact of one or two large economies.
Why it matters
A new framework is only useful if it holds up outside the specific case it was built on. Because the gain in explanatory power appears in every one of the 33 nations studied, not just the four baseline countries examined in closest detail, TMR and CPM function as general-purpose measures rather than a fix tailored to one dataset. That is what justifies proposing them for wider use in fertility research and reporting, alongside or in place of TFR.
The Total Maternal Rate (TMR) is designed to indicate how common motherhood would be if current age-specific first-birth rates remained constant. It is not a direct measurement of the probability that any particular woman will become a mother.
Like the conventional Total Fertility Rate (TFR), TMR is a period measure. It combines birth rates observed among women of different ages in a particular year to produce a hypothetical lifetime measure. It does not follow an actual generation of women throughout their reproductive lives.
This distinction becomes especially important when the timing of motherhood changes. If women begin having their first children earlier, or previously postponed births occur in a concentrated period, TMR can temporarily exceed 100%. Conversely, postponement can depress TMR without necessarily implying an equivalent increase in eventual lifetime childlessness.
Under constant fertility patterns, TMR would itself remain constant. If the prevailing first-birth schedule were sustained over an entire reproductive lifetime and represented a feasible progression into motherhood, TMR would correspond to the proportion of women becoming mothers under that hypothetical regime, and hence to the first parity-progression ratio (PPR1) for a cohort experiencing it.
Consequently, for a hypothetical cohort of childless 15-year-old girls who survive to age 50 and experience these constant age-specific first-birth rates, TMR would equal their lifetime likelihood of becoming mothers. In this sense, TMR provides an early period indicator of motherhood likelihood, without requiring decades to observe completed cohorts.
This assumption is compatible with the conventional Total Fertility Rate (TFR), which similarly assumes constant age-specific fertility rates. Although a constant TFR schedule does not mathematically require a constant TMR, changes in first-birth patterns must remain consistent with the fact that first births precede subsequent births. The more parsimonious interpretation of a constant fertility regime is therefore that TMR and Children per Mother (CPM) also remain constant. Consequently, a TMR exceeding 1 indicates that the assumed first-birth schedule cannot represent a feasible lifetime progression into motherhood, calling into question the feasibility of the corresponding TFR regime.
TMR should therefore be interpreted as an indicator of prevailing first-birth patterns, rather than an unconditional probability or a forecast of completed childlessness. Its value lies in making changes in entry into motherhood visible alongside changes in family size, which the conventional TFR combines into a single figure.
TMR should also be distinguished from the first parity-progression ratio (PPR1). While TMR sums age-specific first-birth rates across all women, PPR1 measures progression from childlessness to motherhood, using women who have not yet become mothers as the population at risk. PPR1 may therefore be more directly suited to estimating eventual motherhood, particularly where sufficient time has elapsed for postponed births to be recuperated. However, calculating age-specific parity progression requires additional information on the proportion of women who remain childless at each age, which is not consistently available across countries and historical periods.
TMR, by contrast, can be calculated directly from first-birth records and provides a period indicator of changes in entry into motherhood as they occur. Under a feasible, stationary fertility regime with consistent age-specific parity distributions, TMR would correspond to the completed PPR1 of a cohort experiencing that regime throughout its reproductive years. As first-birth timing approaches biological limits, however, there is limited empirical evidence to establish which measure provides a more reliable indication of eventual childlessness, given the scarcity of historical examples of such extreme postponement.
Source: Shaw (2025), Scientific Reports, Methods §2 and Supplementary Information S10, Loss of Explanatory Power in TFR (country-level R² gains, reproduced below).
Across Italy, Japan, the United Kingdom, and the United States, Children per Mother (CPM), the average number of children born to women who become mothers, has remained close to constant for four to five decades, even as the national Total Fertility Rate fell substantially (Fig. 1, reproduced below, for the United States). Country means were found to be: Italy 2.10 children (σ=0.18), Japan 2.20 (σ=0.12), United Kingdom 2.34 (σ=0.09), United States 2.48 (σ=0.10) (1970–2020, Italy to 2014; country-by-country distributions shown in Fig. 7, also reproduced below).
Why it matters
Given the finding that mothers are not having fewer children in recent decades, a falling national birth rate cannot be explained by parents having smaller families. The decline must therefore be reflected in a falling period indicator of motherhood likelihood, rather than smaller family sizes among mothers.
Source: Shaw (2025), Scientific Reports, Fig. 1 and Fig. 7 (reproduced below).
TMR did not decline smoothly. Breakpoint analysis identified sharp, sudden falls: in Italy, Japan, and the United Kingdom around 1974; in the United States in 1971, and again in 2008. In Japan, a prefecture-level analysis found TCR (the complement of TMR) rising from 1974 in all 47 prefectures simultaneously, metropolitan, rural, and island regions alike, which a binomial test found statistically improbable by chance (p<0.0001). Monthly birth data narrowed the shift to a cluster of breakpoints beginning October 1974 and peaking in November, consistent with a roughly one-year lag combining time to pregnancy and conception-to-birth (gestation) following the onset of the 1973–74 oil shock in Japanese media on 19 October 1973 (Fig. 8, reproduced below).
Why it matters
A shift occurring across an entire national population, across all settlement types, within a matter of months is more consistent with a shared external shock than with a slow-moving cultural change in preference for having children.
Source: Shaw (2025), Scientific Reports, Fig. 6, Fig. 8, and Supplementary Information S11 (Fig. 8 reproduced below).
Children per Mother (CPM) was found to broadly align with women's stated ideal family size as reported in existing survey literature (illustrated below). This raises the possibility that the gap between intended and achieved fertility is disproportionately concentrated among women who remain childless. This interpretation supports Beaujouan et al.'s finding in the literature that childlessness in twentieth-century Europe was driven largely by structural or circumstantial causes rather than a decline in the desire for children.
Why it matters
Any theory of falling birthrates must now account for two things at once: stable family size among mothers (Finding 3) and a persistent gap between desired and achieved parenthood. These findings highlight the importance of understanding why motherhood rates decline, including the roles of changing intentions, postponement, and circumstances that prevent intended parenthood.
Source: Shaw (2025), Scientific Reports, Discussion section, citing OECD ideal-family-size data and the fertility-gap literature.
The research paper is focused on the novel methods Shaw introduces to demography; to demonstrate their utility, however, these methods are used to explore trends in modern fertility, including the stability of motherhood and declining period motherhood rates, under a new lens. The paper did not set out to demonstrate any specific such pattern, and further research into these respective components of fertility is encouraged.
The paper's central methodological argument is that the Total Fertility Rate (TFR), the most widely used fertility indicator in demography and policy, is a composite measure that obscures more than it reveals. Because TFR is the product of two largely distinct quantities, TMR (an indicator of motherhood likelihood) and CPM (family size among mothers), a single TFR value is compatible with many different underlying combinations of the two. Reporting TFR alone therefore discards information rather than summarising it cleanly.
This is not a qualitative impression; it was measured directly. In the paper's entropy analysis, aggregating TMR and CPM into TFR incurs a 3.279-bit information loss, and TFR retained only 51.1% of the information present in its two components (a 48.9% loss, 95% CI 48.86–50.68%), a loss the paper notes exceeds the theoretical benchmark predicted by Pippenger's Theorem for a measure of this kind. Separately, across 1,470 country-years, each individual TFR value (rounded to one decimal place, as is standard in demographic reporting) corresponded on average to roughly 13 distinct TMR–CPM combinations (Fig. 5, reproduced below).
A concrete example from the paper makes the practical consequence clear. In the United States, the TFR in 2016 was identical to that of 1980, at 1.82. Read as a single number, this would suggest fertility patterns had not changed. But over the same period TMR fell from 76.1% to 69.4%, a decline fully offset, and therefore fully concealed, by a corresponding rise in CPM. Two years with the same TFR described two measurably different societies.
Source: Shaw (2025), Scientific Reports, Results (entropy analysis, TFR-ambiguity simulation and Fig. 5, and the 1980–2016 US comparison).
The paper also distinguishes period measures from the cohort measures more traditionally used in demography. Family size has occasionally been analysed as a cohort measure conditional on motherhood, and childlessness has occasionally been examined retrospectively once a cohort completes its reproductive years. But a period measure of societal childlessness of the kind formalised here, TCR, is, in the paper's own words, “less commonly used in demography than cohort-based alternatives… It has not been formalized as an official indicator until now.” CPM as a standalone period measure is described the same way: family size has “occasionally appeared in fertility research” as a cohort measure, but “the societal period measure Children per Mother (CPM) has not previously been formalized as a standalone indicator.” The practical difference matters: because cohort measures are retrospective by design, a shift in childbearing patterns is often not visible in cohort data until decades after it began, whereas the period measures TMR and CPM can register it as it happens.
The paper illustrates how this reporting would work in practice with recent, real-world data for ten countries, shown in Table 1 below: TMR and TCR alongside CPM and the conventional TFR figure, for the most recent year available in each country. Reported side by side rather than collapsed into TFR alone, the table shows at a glance how far apart the incidence of motherhood (TMR) is across countries even where TFR values are similar, and how consistently CPM sits in a narrower band than either TFR or TMR.
Taken together, the microdemographic decomposition (TFR = TMR × CPM) supplies direct empirical evidence from birth records that declining period motherhood rates, rather than smaller family sizes among mothers, account for the long-term decline in fertility in the countries studied. Children per Mother (CPM) remained stable for four to five decades while national Total Fertility Rates fell substantially (Finding 3); because CPM held close to constant over this period, the fall in Total Maternal Rate (TMR) is, as a direct arithmetic consequence of the TFR = TMR × CPM identity, responsible for virtually all of the observed decline in TFR (Finding 1). Several of the TMR declines were abrupt rather than gradual, and in the Japanese case occurred simultaneously across all 47 prefectures, a pattern more consistent with a shared external shock than with a slow-moving cultural change in preferences (Finding 4). CPM also broadly aligns with women's stated ideal family size in the survey literature, raising the possibility that the fertility gap is disproportionately concentrated among women who remain childless (Finding 5).
These findings are consistent with the possibility that unplanned or circumstantial childlessness contributes to declining motherhood rates, in line with the hedged causal language discussed below. It is important to be precise about what the birth-record analysis does and does not show: none of its empirical patterns, by themselves, demonstrate that voluntary childlessness has risen as the primary driver of falling fertility. Formally distinguishing voluntary from unplanned childlessness requires intention data of some kind; registration-based measures (TMR and TCR) alone cannot make that separation precisely. As this page discusses elsewhere, survey-based intention data carry their own well-documented measurement problems. No single data source currently settles this question cleanly, but the findings of this paper describe a structural pattern in shifts in childlessness that any proposed theory of intention would have to align with.
The paper is explicit about the limits of its own causal claim. In its own words: “Although this study does not claim a definitive causal relationship between economic shocks and fertility outcomes, the findings nonetheless offer compelling evidence of a strong association.”
The paper also notes that TCR “encompasses voluntary, medically and biologically involuntary, as well as unplanned (circumstantial) childlessness,” and concludes that the presence of rapid shifts during periods of economic uncertainty suggest unplanned childlessness may be a notable contributor, language of association and hypothesis, not of a settled causal finding.
The shock timing itself is documented directly in the paper: Italy, Japan, and the United Kingdom around 1974, coincident with the 1973–74 oil crisis; the United States in 1971 and 2008, the latter coincident with the 2007–08 financial crisis.
Evidence Clarifications
Since the film's release, criticism has been published by several organisations and commentators, most substantially by BirthgapFacts.org, run by Population Balance, an activist organisation whose stated goal is to shrink the human population. The points below are addressed because they can be checked directly against a primary source in a few minutes, independent of any view on the wider argument.
The quotation “I think that the biggest problem the world will face in 20 years is population collapse” is presented in published criticism as a statement by the filmmaker. In the documentary, it is archival footage of Elon Musk, captioned on screen with his name, appearing at 00:00:40.
This clip sits within an opening sequence that intercuts archival footage of contrasting public perspectives on population. Paul Ehrlich, author of The Population Bomb, who held the opposite concern (overpopulation rather than population decline), appears in the same sequence at 00:00:07 and 00:00:45, either side of the Musk clip, and again later at 01:41:12. The juxtaposition of Musk and Ehrlich, holding opposite views, supports reading the sequence as a presentation of contrasting third-party perspectives rather than the filmmaker's own position.
Source: Birthgap, youtu.be/m2GeVG0XYTc, timecodes as above.
The passage “We can sustain a much larger population if the materials that we consume are both produced sustainably and we consume them in moderation… if the world's population gets to 10 billion, there is hope that we could have a sustainable planet, yes” is presented in published criticism as a statement by the filmmaker.
Every released version of Birthgap has been checked, and the filmmaker is not the source of this passage in any of them. Janet Ranganathan, then Vice President for Science and Research at the World Resources Institute, appears elsewhere in the released feature-length film as an interviewee, but the specific exchange quoted above does not appear in the free theatrical/YouTube release. It does appear in the paid Extended Version of the film, Chapter 7, at timecode 1:39:15, where it is unambiguously an on-camera answer from Ranganathan to a question put by the filmmaker, not a statement by the filmmaker himself.
Source: Birthgap, Extended Version, Ch. 7, 1:39:15, available via Patreon (paid content).
Published criticism states that “56% of childless adults just don't want children.” In Pew Research Center's 2021 survey of US adults aged 18–49 without children, 44% said it was not too or not at all likely they would have children someday; only that 44% was then asked why, and of those, 56% cited not wanting children. 56% of 44% is approximately 25% of childless adults under 50, not 56%.
A more informative figure comes from Pew's own 2024 survey of adults aged 50 and older who never had children, a group whose childbearing years are complete. Among this group, the most commonly cited major reason for not having children is that “it just never happened” (39%), ahead of “didn't find the right partner” (33%) and “just didn't want to have children” (31%). Only 32% say they never wanted children at all.
Source: Pew Research Center, 2021 and 2024 surveys.
Published criticism cites a 2023 Michigan State University study reporting that 20.9% of adults are “childless by choice,” building on an earlier 2022 study by the same authors, as evidence that voluntary childlessness is common. Both studies are peer-reviewed and were conducted carefully; the concern here is about how the finding is applied, not about the research itself.
The Michigan sample spans adults aged 18 to 80, a range covering people who formed their families across markedly different decades, economic conditions, and social norms. The authors are explicit in both papers that findings from a single US state should not be generalised without national replication, and that the study does not capture respondents' reasons for not wanting children.
By contrast, the National Survey of Family Growth (NSFG), a nationally representative US survey, places voluntary childlessness among women aged 15–49 at 7.6% (2015–2017) and 9.8% (2017–2019), a narrower and more recent age band, and a figure roughly a third to a half the size of the Michigan estimate.
Why it matters: a regional, multi-decade sample and a nationally representative, current-cohort sample are measuring different things, even when both are conducted well. Citing the higher figure without this context overstates how common voluntary childlessness is shown to be among today's childbearing-age population.
Published criticism cites NSFG data (Martinez et al., 2006–2010) reporting that 72% of permanently childless women were voluntarily childless, presented as evidence that voluntary childlessness dominates. This figure is real, but it describes a much narrower group than “childless women” in general, and presenting it without that context creates the same kind of misleading impression as the Pew 56% figure above.
The NSFG's own current key statistics illustrate why. Among all women aged 15–49, only 10.9% (2015–2017) to 13.0% (2017–2019) are permanently childless, voluntarily or for medical/biological reasons combined. The remaining 31.0–31.8% of all women who are childless at the time of the survey still expect to have children, and are not counted as permanently childless at all. Within the small permanently-childless group specifically, the voluntary share runs to roughly 70–75% in this more recent NSFG data (7.6/10.9 and 9.8/13.0 respectively), broadly consistent with the 72% figure cited from the earlier 2006–2010 data, once the same restriction is applied.
Why it matters: permanently childless women are themselves a small fraction, roughly one in nine to one in eight, of all women aged 15–49. The great majority of childless women in any given NSFG wave still expect to become mothers. Citing the 72% figure without stating what share of all childless women it covers overstates how common voluntary, settled childlessness is among women broadly.
Independent demographic research supports the qualitative core of the unplanned-childlessness argument, distinct from the disputed “10–10–80” numeric heuristic addressed below. Keizer, Dykstra & Jansen (2008), drawing on the Netherlands Kinship Panel Study, find that only a small group of people, so-called “early deciders”, state at an early age that they do not intend to have children, while a larger group “ends up childless without having explicitly pondered the decision.” They argue it is therefore more accurate to describe this as remaining childless than choosing childlessness: “the outcome of never having made the decision to actually have children,” typically following choices to delay marriage and prioritise education or career rather than any decision about parenthood itself.
Shaw's findings sit alongside a wider and growing body of institutional work on unmet fertility intentions. UNFPA's State of World Population Report 2025, titled The Real Fertility Crisis: The Pursuit of Reproductive Agency in a Changing World, documents a persistent gap between desired and achieved childbearing.
A more recent and directly relevant figure comes from UNFPA's Lives, Choices and Futures report, based on the 2025–2026 Demographic Futures Survey of over 108,000 internet-connected adults aged 18–39 across 73 countries. Among respondents aged 35–39 in the survey's Western Europe, Australia, Canada, and United States regional grouping – the group of countries closest to those studied in Shaw (2025) – 49% of men and 41% of women had no children. Within that childless group specifically, 73% of men and 62% of women said they would ideally like to have children. The report itself cautions that, while weighted by age, sex, and education, the sample is drawn from internet-connected respondents and is not nationally representative; it should be read as indicative of this population rather than as a population-level estimate.
Separately, the US National Survey of Family Growth (NSFG) has tracked fertility-intention measures across every survey cycle since 1973, with response rates historically ranging from 63.4% to 90.2%. In its 2022–2023 redesign the survey moved to a web-first self-completion design, with in-person interviewer follow-up offered only to non-respondents; the response rate fell to 26.8%, the sharpest single-cycle decline in the survey's fifty-year history.
Under this redesign, the weighted share of women reporting they have or want children fell to 77.6% among web respondents, against 88.4% among the concurrent, same-wave in-person respondents, close to the 90.0% recorded in the prior, fully in-person 2017–2019 wave. The two groups answered the same questions in the same survey cycle, although differences in respondent composition may also have contributed to the results.
A further check does not depend on any attitudinal question at all. Whether a respondent already has a child is a fixed biographical fact, not an intention, and can be checked against independent Census/CPS benchmarks. Under the web mode, the reported share of women who are already mothers fell to 42.2%, below every prior NSFG wave and below the concurrent in-person figure (51.4%) from the same cycle, a share of already-realised motherhood that should not change simply because respondents complete a survey online rather than in person. This pattern is consistent with differences in survey response or measurement, rather than with a genuine and sudden shift in underlying fertility intentions or family circumstances.
Source: Shaw, S.J. (2026), unpublished manuscript, not yet peer-reviewed. This section will be updated to a preprint or journal reference once one is available.
Brief definitions of the fertility measures referenced on this page, drawn from Shaw (2025), Scientific Reports, Supplementary Information.
TMR and TCR describe hypothetical outcomes under prevailing fertility patterns, rather than the observed lifetime outcomes of actual generations. Because they are calculated from annual birth rates, these indicators can occasionally fall outside the 0–100% range.
If anything on this page is inaccurate, we want to know. Email comments@birthgap.org and any confirmed error will be corrected here, with the change and its date recorded below.
No corrections to date.
