Fertility in a Pandemic: Evidence from California

Abstract

The COVID-19 pandemic was accompanied by social and economic changes previously associated with fertility delay and reduction, sparking widespread discussion of a “baby bust” in the U.S. We examine fertility trends using restricted vital statistics data from California, a diverse population of 40 million contributing 12% of U.S. births. Using time series models that account for longer-run fertility trends, we observe modest, short-term reductions in births from mid 2020 through early 2021. Birth counts in subsequent months matched or even eased the pace of fertility decline since the 2008 recession and are unlikely a function of the pandemic alone. Responses to the pandemic were heterogeneous. Fertility declined markedly among the foreign-born population, largely driven by changes in net migration. Among the U.S.-born population, the short-term pandemic-attributable reductions were largest among older, highly educated people, suggesting mechanisms of fertility reduction disparately accessible to those with the most resources. We find no evidence of a strong population fertility response to the pandemic’s accompanying employment shock, providing additional evidence of a growing divide between macroeconomic conditions and fertility patterns in the U.S.

Population fertility is regularly responsive to contextual conditions—including large-scale economic, social, political, and environmental change (Karaman Örsal and Goldstein 2018; Seltzer 2019). These patterns interest social scientists because they offer one window on central questions about why people have children, and when people have children. Understanding the relationship between macro-contextual conditions and population fertility is necessary for the development of population-related policy, and important for forecasts of future population trends.

The COVID-19 pandemic presented an unusual challenge for population scientists predicting demographic change; multiple proximate causes of fertility shifted nearly simultaneously. The pandemic was initially accompanied by social and economic changes associated with fertility reductions—including spiking unemployment, worsening physical and mental health, and reduction of in-person interaction of non-cohabiting people—seeding widespread discussion of a looming “baby bust.” Indeed, early evidence of pandemic-associated aggregate fertility declines in late 2020 and early 2021 was recorded for multiple European populations and multiple U.S. states (Aassve et al. 2021; Cohen 2021).

At the same time, the classification of abortion as an elective medical procedure in several states reduced abortion rates (Jones, Lindberg, and Witwer 2020; White et al. 2021) and delays in access to regular medical care made contraception more difficult to obtain in the early months of the pandemic (Lindberg, Bell, and Kantor 2020). The pandemic also arrived during a period of sustained fertility delay in the U.S.; any pandemic-related downward pressure on fertility may have been muted for older individuals who had already delayed entry to childbearing and may have been hesitant to delay further, given their age.

The relevance of these pandemic-induced changes was not experienced equally across U.S. families. The unequal burden of the pandemic’s economic and health effects is now well-documented. Black and Hispanic people, people with less education, and those living in communities with fewer resources have disproportionately borne COVID-19 infection and mortality risk, unemployment, food insecurity, medical care reductions, incomplete childcare, and the accompanying stress of these changes (Hooper, Nápoles, and Pérez-Stable 2020; Page et al. 2020; Torche and Nobles 2022; Verdery et al. 2020). To the extent that the pandemic reduced fertility demand and potentially reduced fecundity (people’s biological capacity to achieve childbirth–e.g., Nobles and Hamoudi 2019; Wilcox 2010), these effects may have disproportionately occurred among Black and Hispanic people and among economically less-resourced people (Cohen 2021).

Providing timely evidence on population fertility is difficult because of the timing lag created both by the length of pregnancy and by the hurdles of releasing administrative data. U.S. vital statistics are >99% complete and are unambiguously the highest quality source of information on birth counts. The coverage rate is particularly important for the study of population variability in fertility patterns, because unlike surveys with incomplete coverage, the most disadvantaged people are captured in the vital statistics data. Detailed administrative records, however, are typically only released annually, with an associated 9–18 month lag, and with considerable coarsening of key measures like location and birth timing.

To provide timely evidence on fertility change, we acquired early-release vital statistics data on births from the Department of Public Health for the state of California from January 2014-November 2022. We use the data to ask (a) whether the pandemic was associated with reductions in population fertility, (b) for how long, and (c) for which population subgroups. In addition to providing conclusive evidence contradicting a “baby bust” of any consequential size, the results provide new evidence of a striking age and socioeconomic gradient in the fertility response to the pandemic. Using data on ART-assisted births, we argue that the fertility patterns documented here likely reflect intentional fertility delay that was differentially achieved by the most socioeconomically-resourced individuals. We conclude by discussing several implications for studying population welfare in the context of the pandemic and longer-run fertility change in the U.S.

Theoretical Predictions of Changes in Fertility Timing

Population science provides a rich set of models to theorize the expected fertility response to the emergence of a large-scale health threat like COVID-19 (Aasve et al. 2020). Most models of fertility describe individuals or couples seeking, delaying, or avoiding a next birth by intentionally or automatically engaging in reproductive behaviors, including partnering, sex, contraception, and abortion (Bongaarts and Potter 1983). People’s desire to seek a next birth is shaped by processes operating at multiple levels: features of the individual and couple, their household, and the community in which they live—like the quality, status, and duration of partnerships, individual and partner ages, economic circumstances and outlook, the local cost and feasibility of raising children, and attitudes about childbearing in individual’s kin and social networks (Bongaarts 2015; Hayford 2009; Hotz, Klerman, and Willis 1997). The ability to achieve birth—or its avoidance—is in turn shaped by an array of constraints, including the availability of partners, physical and mental health, and access to contraception, abortion, and reproductive health care.

Fertility timing, or when individuals have offspring, is shaped by similar processes but may be particularly sensitive to people’s assessment of changing circumstances and outlook on future circumstances (Guetto et al. 2022; Guzzo 2022; Trinitapoli and Yeatman 2011)—e.g., the desire to avoid unsafe conditions or to align births with specific life events. Because the probability of conception in any given month is low even for healthy couples (Wilcox 2010), fertility timing also includes a considerable stochastic component.

These elements are all relevant in the context of large-scale social, economic, and health events—including a pandemic accompanied by economic decline and changes in social interaction. In many contemporary societies, population fertility and economic conditions are procyclical (Karaman Örsal and Goldstein 2018; Seltzer 2019). In the context of rising unemployment, we would typically expect reductions in overall fertility rates, particularly if men’s employment and expected wages decline relative to women’s, increasing the opportunity cost of childbearing to different-sex couples (Siegel 2017). Other factors that accompany pandemics can contribute to population fertility decline. Disruptions in people’s ability to meet new partners during quarantines or other social-distancing policies mechanically result in temporal fertility delay (Wagner, Choi, and Cohen 2020). Reductions in health—either from acquiring viral infection or associated with the physiological stress of unemployment and uncertainty—can reduce fecundity (Catalano et al. 2005, 2022; Chen et al. 2022).

Other factors put upward pressure on fertility. Changes in medical care access, including reductions in access to abortion and contraception, would both be expected to increase unplanned fertility, all else equal (Bailey, Bart, and Lang 2022). Though currently less salient in the context of COVID-19: pandemics that increase mortality risks for fetuses and infants (such as Zika, HIV) or for reproductive-age adults (such as HIV and some forms of influenza), can increase period and cohort fertility (Hayford, Agadjanian, and Luz 2012; Preston and others 1978; Trinitapoli and Yeatman 2011). Early-life and reproductive age mortality effects are an important source of fertility change in other pandemics; we would expect these to be much smaller (and potentially non-existent) in the context of COVID-19 given the concentration of mortality risk among older adults.

Population Variation in Predicted Fertility Reponses to Pandemic Exposure

Most models of reproductive behavior predict socioeconomically heterogeneous fertility responses to an emerging pandemic. That is, in the context of a new health threat, people with economic resources are better able to access mechanisms of pandemic avoidance—like private transportation or remote work, and are able to more quickly access mitigation resources, like vaccination or antiretroviral treatment (Link and Phelan 1995; Phelan, Link, and Tehranifar 2010). To the extent that unemployment increases economic vulnerability among the least resourced people and reinforced perceptions of economic despair, we might also expect small increases in fertility among disadvantaged young people seeking purpose in parenting (Kearney and Levine 2014).

As a result, the pandemic-driven mechanisms of fertility change described above, such as reductions in health and increases in economic insecurity, would all be expected to disproportionately affect people with lower socioeconomic status. To the extent that these exposures reduce both fertility and fecundity, we would expect to see larger fertility reductions for less educated individuals and people living in less wealthy communities—all else equal. However, socioeconomic status also shapes access to the resources that allow people to achieve fertility delay, such as contraception and abortion and the time to implement them effectively. As a result, downward pressure on population fertility from people’s desire to delay births may be offset, or even numerically dominated, by increases in unplanned pregnancies (Bailey, Bart, and Lang 2022).

A relevant source of socioeconomic heterogeneity in response arises when a pandemic occurs in the context of declining population fertility (Aasve et al. 2020). When the COVID-19 pandemic reached the U.S., the population had experienced over a decade of fertility decline (Hartnett and Gemmill 2020; Osterman et al. 2022), as people delayed fertility to older ages. Fertility delay is most pronounced among the highest educated; we might predict that this group, many of whom will have already delayed fertility into their thirties, to be less likely to time fertility in response to an emerging pandemic. Age constraints may make further delays untenable.

In the U.S., population heterogeneity would also be expected to occur by race and nativity. For example, Pirtle (2020) develops a model of racial capitalism in which the structural conditions that predate pandemics and that shape their disparate impact on racialized communities result in much larger, more deleterious exposure to the morbidity and mortality accompanying emerging viral threats. In the U.S., we would expect that the mechanisms that promote desire for fertility delay—and also the difficulty in achieving it—would disparately affect Black, Latino, and immigrant families.

Evidence on Pandemic-Driven Fertility Change

Though the U.S. has not experienced a pandemic with a health, mortality, or economic impact the size of COVID-19 in many decades, evidence of fertility in the context of other emerging infectious diseases is consistent with several of the theoretical predictions described above. When the Zika virus reached Brazil in 2015 with demonstrated health consequences for pregnant people and fetuses, Brazilian women with both high and low socioeconomic status (SES) expressed a desire to delay fertility, but only economically resourced people were able to take advantage of multiple strategies to reduce exposure to the virus. High SES women also expressed disparate confidence in their ability to achieve goals of delayed fertility (Marteleto et al. 2017). Ultimately both more- and less-educated women reduced childbearing, but the reductions were substantially larger for more-educated women (Rangel, Nobles, and Hamoudi 2020). In a much earlier example: following the 1918 influenza pandemic, U.S. births declined significantly 9–10 months following peak influenza mortality rates (Chandra et al. 2018). Though fertility rates by parental education are difficult to estimate in the early 1900s, the resulting 1919 birth cohort was distinctive from other cohorts in the 1912–1922 decade: children were more likely to be born to illiterate fathers and fathers with lower status occupations (Beach et al. 2022).

The COVID pandemic is distinctive because the mortality risks to fetuses and pregnant people were not nearly as high as during the Spanish flu or Zika, though COVID-19 infection in pregnancy does carry morbidity risks for mothers and infants (Thoma and Declercq 2022; Torche and Nobles 2022). Fertility delay in the context of the COVID-19 pandemic would have likely been a response to a bundled set of conditions, including economic, housing, and health uncertainty. Indeed, multiple studies suggest that people desired to avoid pregnancy in the initial months of the pandemic. Lindberg (2020) found that though both higher and lower SES people expressed a desire to delay fertility, a larger share of less-resourced people expressed such desire. In a longitudinal study of fertility desires, Koenig and colleagues (2022) found that younger, Black and Latina respondents were more likely than older or white respondents to desire fertility delay when initially interviewed in 2020. These same groups were also more likely to express a desire for continued delay a year later. Respondents who had experienced economic loss during COVID-19 were also more likely to express desire for delayed fertility.

Multiple researchers initially predicted a dramatic decline in fertility after the onset of the COVID-19 pandemic. One study estimated 300,000 to 500,000 fewer children resulting from the public health crisis and ensuing economic recession (Kearney and Levine 2020a, 2020b). As we now know, a reduction of this scale did not materialize. Using national birth data from 2016 to 2021, Kearney and Levine (2022) found a short-term decline in births from October 2020 to February 2021 followed by a rebound from March to December 2021. The net result of these opposing trends is a very small reduction in fertility compared with the expected number of births obtained from pre-pandemic trends. The authors speculate that the reduction in births during 2020 was driven by decisions to postpone or avert fertility in the context of economic decline. Bailey et al. (2022) found that the decline in births during 2020 occurred too early in the pandemic to be driven entirely by fertility adjustments, and the reductions were substantively larger for the foreign-born population, consistent with reduced entry of foreign-born people into the U.S. during the pandemic. The US-born population experienced a much smaller decline in fertility in 2020 followed by an increase in 2021. The observed increase in fertility among the native-born was more pronounced among distinct subpopulations, those younger than 25, those age 30–34, those with first births, and the college educated.

We extend this line of work by considering multiple factors that may have contributed to fertility changes over the course of the pandemic. We assess evidence of behavioral adaptations to a changing pandemic environment and consider constraints emerging from economic and institutional factors, such as an early-pandemic spike in unemployment and restricted access to assisted reproductive technologies. Given concerns about appropriate inference in the context of a multi-year fertility decline in the U.S. (see, e.g., Gemmill et al. 2022), we use time-series models to identify the impact of the pandemic net from trend, seasonality, and other sources of temporal autocorrelation in birth data.

The COVID-19 Pandemic in California

California is the most populous state in the United States, comprising 12 percent of both the total population and of all births in the United States. The state is exceptionally diverse in terms of race/ethnicity and socioeconomic standing of its population. In 2020, no single racial/ethnic group constitutes a majority in the state. Approximately 39 percent of the population was Hispanic, 35 percent was white, 15 percent was Asian or Pacific Islander, 5 percent was Black, 4 percent was multiracial and less than 1 percent was Native American/Alaskan Native. Based on the official poverty rate, California is comparable to the national average in 2020 (12.6 percent of households compared to 13.2 percent) but its level of income inequality is among the highest in the nation.

California was one of the first states to respond to the emergent evidence of an epidemic in 2020. As reported by Catalano et al. (2021), large employers, primarily in the technology sector, began telling their California workers to stay at home effective March 8, 2020. Counties in the San Francisco Bay Area issued stay-at-home orders effective March 17, and the Governor issued similar orders for the remainder of the state on March 19.

Concern about the new epidemic emerged prior to these institutional responses, given the early exposure of the state to the virus. Ten of the first twenty confirmed COVID cases in the United States happened in California, with the first case confirmed on January 26, 2020. While concern about the threat posed by the virus may have emerged as early as December 31, 2019, when the WHO announced that a mysterious pneumonia was affecting dozens of people in Wuhan, China, the early direct exposure to the virus in California suggest widespread awareness and concern among the California population as early as in February of 2020.

In April 2020, the unemployment rate more than tripled, exceeding 15% for the population, and 17.5% for Black Californians specifically (U.S. Bureau of Labor Statistics 2023). Economic recovery began shortly in most parts of the state, and unemployment had fallen to 6.5% by December 2020 and returned to pre-pandemic levels around 4.3% in the Spring of 2022. During the initial months of the pandemic, a number of medical procedures were delayed to reduce infection spread and to free up medical staff and equipment. Most clinics offering assisted reproductive technology (ART) closed in March 2020 and remained closed for 6–12 weeks. Nationwide, the number of births supported by ART clinics fell by 4.7% in 2020 relative to 2019 (ASRM/SART 2022).

Abortion and contraceptive services were less disrupted in California than they were in some other parts of the U.S.—e.g., states that classified abortions as elective and halted their performance in the early months of the pandemic (White et al. 2021), or states that did not mandate reimbursement parity for telehealth contraceptive services (Ellison, Cole, and Thompson 2022). Nevertheless, the pandemic occurred amidst ongoing reductions in funding for reproductive health services around the country, and this increased the proportion of Californians lacking access to affordable contraceptive care (Smith et al. 2022).

These features of the COVID-19 pandemic have testable implications for patterns in population fertility. If people sought to avoid pregnancy in the context of the uncertainty brought about by the early months of the pandemic, we would expect to see population fertility decline by late 2020. If the subsequent unemployment shock exacerbated desire to delay fertility, we would expect to see further fertility reductions continue or even grow into Spring 2021, and we would expect these to be clustered among people working in sectors experiencing the higher levels of unemployment, including younger people and people working in positions that require fewer years of formal education.

As in other parts of the country, the COVID-19 pandemic occurred in the context of long-run fertility decline. Figure 1 plots monthly birth counts from January 2014 through December 2021. Birth counts had declined from about 42,000 births per month in 2014 to about 37,000 births per month by 2019, prior to the start of the pandemic. Strong seasonality in births was also evident. As in other U.S. states, there are roughly 16% fewer births each February relative to births each August. In light of both of these patterns, any attempt to model the effects of the pandemic’s onset on fertility change requires an approach that accounts for trends, seasonal shifts, and other sources of temporal autocorrelation (Gemmill et al. 2022). Ignoring these patterns runs the risk of attributing reductions in fertility in the Winter of 2020–2021 to the pandemic, when long run changes and seasonal fluctuations are the driving factors. We use such an approach here.

Figure 1. Births in California by month, 2014 – November 2022.

Source: California Department of Public Health. Red line indicates the onset of the pandemic in March 2020 (red vertical line). Smoothed nonlinear trend line depicted in black.

METHODS

Data.

We use georeferenced vital statistics data for births occurring in California between January 1, 2014 – November 30, 2022. The California Department of Public Health collects information about all births occurring in the state as well as births to California residents occurring in other states. Birth records contain detailed information about the timing and location of delivery, the medical procedures surrounding the delivery, information on birthing persons’ parity, age, race, ethnicity, education, and zip code of residence. Records also document the use of assisted reproductive technology to achieve the pregnancy. The education of the person giving birth provides an important measure of socioeconomic heterogeneity in the sample. Residential zip code also facilitates measurement of area-level variation in socioeconomic resources given substantial socioeconomic segregation across the state. To create this measure, we use an index created with information on the following measures collected from the Census Bureau’s 2015–2019 American Community Survey (ACS) data: proportion of adult residents in the zip code with a high school diploma or less, zip code median household income, proportion Hispanic, proportion Black, Gini index, proportion foreign-born, proportion households with income below the poverty line, and proportion individuals without health insurance. We combined these indicators using a principal component analysis. We extracted the first component form the PCA and divided it into quartiles to index area-level socioeconomic variation.

Approach: Detecting COVID-associated changes in fertility.

We test for COVID-associated shifts in total population fertility with Autoregressive Integrated Moving Average (ARIMA) models that use information on the pre-pandemic period: January 2014 – February 2020 to forecast expected fertility during the March 2020 – November 2022 period, and test for deviation from these forecasted values. To conduct ARIMA analysis, we create aggregate monthly time series datasets consisting of birth counts for the entire state population, and then stratified across population subgroups. We then divide all values by the mean pre-pandemic month birth count; analysis of this transformed data facilitates interpretation of the results as proportional deviations from expected values.

Time series data can contain sources of temporal autocorrelation including trend, seasonality and the tendency for high and low values to persist over several time periods, which can confound the effect of exposure (Catalano et al. 2008). ARIMA fits univariate models where the disturbances are allowed to follow a linear autoregressive moving-average specification. The ARIMA models allow for seasonal parameters to account for the marked seasonality of number of births (Bobak and Gjonca 2001; Darrow et al. 2009). Model selection is based on evaluation of the autocorrelation and partial autocorrelation plots (Pankratz 1983, 2012), assessment of the model’s goodness-of-fit based on the Akaike Information Criterion (AIC) and evaluation of the residuals of the ARIMA model by means of Portmanteau and the Bartlett’s periodogram-based tests to ensure that they are white noise. We introduce a set of indicator variables to the model identifying post-treatment months between March 2020 and November 2022. Parameter estimates associated with these indicator variables capture the departure in observed birth counts after the onset of the pandemic from birth counts expected based on past observations in the time series. We examined birth counts in their original metric as well as in logged form and found similar results across specifications.

The general logic of the ARIMA approach is similar to other regression methods: we use past trends to generate approximations of the counterfactual birth counts each month – i.e., what might have been expected if birth counts in 2020–2022 were not influenced by COVID? – and test for deviation from that counterfactual. The value of ARIMA modeling is that it addresses autocorrelation in the patterning of outcomes that is not captured with parsimonious parameterizations of time–a linear or quadratic time trend, for example. Gemmill and colleagues (2022) have demonstrated how consequential for inference it can be to ignore these forms of autocorrelation in analysis of time series data. In the current study, tests of residuals all indicated the importance of relying on ARIMA specifications for inference. However, to make the results as accessible as possible, we have also estimated a set of constrained linear regressions that use calendar month fixed effects and linear time trends fit to the pre-pandemic period and use these to estimate expected values in March 2020-November 2022 and test for deviations. These results are shown in Appendix A and lead to results that resemble the preferred ARIMA specifications in some instances and which depart to our preferred models in other instances.

The timing of change in birth counts is important: reductions in birth counts in approximately the first 6 months after the pandemic began will most likely operate through the departure of pregnant people from the population or through miscarriage. Later reductions could include shifts in the biological aspects of fertility–i.e., fecundity or the ability to get pregnant and carry to term–as well as fertility behaviors related to intercourse and contraception. Declines right after the normative duration of pregnancy since onset of the pandemic –approximately in December 2020– will signal a response to the novel stressor driven by biological or behavioral factors. In addition, the presence or absence of temporal discontinuities in births conceived via assisted reproductive technology allow us to gauge the contribution of restricted medical services in the early months of the pandemic to changing fertility patterns.

Based on the theoretical models of fertility change in the context of wide-scale societal disruption, and empirical evidence on differences in the pandemic’s effects on infection risk and economic disruption by socioeconomic advantage and race and ethnicity in the U.S., we might expect larger reductions in fertility for people who had higher risk of reductions in health and employment–lower SES people and people from racially minoritized populations. Alternatively, it is possible that more resourced people were more able to achieve fertility delay. We would also expect that older people may be less responsive to the pandemic’s arrival because age may make fertility delay difficult.

To test for these explanations, we stratify the sample by the birthing person’s age, race and ethnicity, foreign-born status, education, and zip-code level socioeconomic status. We extend the analysis and estimate ARIMA models on each of these subgroups. Tests stratified by age and by parity are also informative about whether evidence of fertility decline may be indicative of fertility delay (a “tempo” shift) or plausibly reflects change that are likely to become a reduction in cohort completed fertility (a “quantum” shift). Decline in fertility at older ages, for example, is less likely to be recouped later on, as older people experience reductions in fecundity with age. Delays at these ages are more likely to lead to reductions in completed fertility. Similarly, reductions in higher-order (3+) births are less likely to be recouped later. If the share of people who desire to be childfree is not increasing, reductions in first births are most likely to be indicative of delay.

Birth counts provide a useful fertility metric because they are well-measured in places with complete vital event registration systems, like the U.S. However, changes in birth counts may reflect underlying shifts in the population at risk of giving birth. As a tool to study population fertility, time series estimates of birth counts assume that net migration among reproductive-age people between the ages of 15–44 years into California is not changing. We test the sensitivity of the results to pandemic shifts in net migration with data from the monthly Current Population Survey (CPS). We generate monthly counts of reproductive-age women in California through November 2022, and use these alongside birth counts to calculate monthly general fertility rate (GFR) values, i.e., the number of live births per 1,000 females of childbearing age between the ages of 15–44 years, first for the full population and then for population subgroups. It is important to emphasize that the value of these results is dependent on the quality of CPS data to capture short-term fluctuations in size and composition of the entire population as well as specific subgroups–we return to this issue in the final section of this study.

RESULTS

Figure 2 plots ARIMA estimates by month for all births in California in blue dots, with 95% confidence intervals. The estimates are interpreted as the proportional deviation of birth counts in each month relative to the predicted value for that month based on trends and seasonal patterns observed pre-pandemic.

Figure 2. Proportional deviation in California birth counts by month: All births, and births without assisted reproductive technology, March 2020 – November 2022.

Note: Estimates from ARIMA models analyzing monthly birth count values divided by the mean pre-pandemic (January 2014 – February 2020) monthly birth count. Estimates indicate the proportional deviation of birth counts in each month relative to the predicted value for the month based on trend, and seasonal patterns, and other sources of temporal autocorrelation observed pre-pandemic. Navy closed circles denote estimates for all births; red open circles denote estimates for births not supported by ART. 95% confidence intervals shown.

We observe birth counts that are 5 to 6 percent lower than expected from August through November 2020. In December 2020, the counts fall to 7.7% below expected values and to 8.2% in January 2021. By March of 2021, birth counts return to expected values. Note that the counterfactual for these estimates are counts that would be expected under the assumption that the pre-pandemic reductions in fertility continued after the arrival of COVID. A return to expected values here implies birth counts in 2021 that fall below pre-pandemic birth counts, because this pattern is predicted by ongoing delays and declines in population fertility that began in the late 2000s and appear to continue into the present period. Some indication of an increase in the number of births in late 2021 is observed, with counts surpassing predicted trends by 4.3% in December 2021. These increases are not sustained over time and provide limited evidence of a fertility rebound at the population level. As we examine in detail later, increases in birth counts in 2021 are heterogeneous across groups, with some subpopulations experiencing substantial gains while other groups showing little departure from counterfactual trends.

Additional analysis helps inform the interpretation of the timing patterns in Figure 2. The below-expected birth counts in Fall 2020 cannot be driven by changes in fertility behaviors after the onset of the pandemic because these pregnancies were already in progress. They are also unlikely to be driven by changes to miscarriage or abortion–below expected values in August 2020 would only be possible if exposures in March and April 2020 were contributing to excess termination of second-trimester pregnancies, a period during which spontaneous and induced abortion are rare–far too rare to account for more than 5% drop in birth counts.

Instead, the reduction in birth counts in late Summer and Fall 2020 is likely related to population change driven by migration patterns. Pre-pandemic, approximately one-third of California births occurred to foreign-born people. It is well documented that the foreign-born population declined sharply after the onset of the pandemic due to the interruption of international travel and the decline in processing of student, scholar, and professional visas. U.S. immigrant visas issued for permanent and temporary residence fell by 48 and 54 percent respectively between 2019 and 2020 (Gelatt and Chisti 2022). The large decline in foreign-born population was strongest for people of reproductive age and was particularly pronounced in California compared to other states (Peri and Zaiour 2023).

In contrast, the contribution of internal migration to population decline appears to be unsubstantial. Studies of interstate migratory flows indicate very minor change compared to pre-pandemic trends, and the continuation of a long-term process of population loss driven mostly by the high cost of living in many state localities. Limited change from pre-pandemic trends is documented by analysis of moving companies (Haslag and Weagley 2022) (Lavelle and Kepner 2022, Haslag and Weagley 2022), of US Postal Service data (Kolko, Badger, and Bui 2021), and of California Department of Finance data (State of California 2022). Based on the birth certificate data, we can directly examine the number of births to California residents occurring in other states as well as the number of births to residents of other states occurring in California. We find that internal migration was indeed a negligible driver of changes in birth counts: The share of out-of-state births to California residents changed minimally from 0.29% in 2019, to 0.30% in 2020 and 0.32% in 2021. Similarly, the share of California births to out-of-state residents was 0.26% in 2019, 0.23% in 2020 and 0.26% in 2021.

To examine the potential influence of these population changes, we use monthly CPS data to generate California population counts and estimate ARIMA models predicting the general fertility rate instead of birth counts (Appendix Figure B2). Doing so reveals similar, if weaker patterns: we observe a 6% reduction in the general fertility rate (GFR) in December 2020 - January 2021, followed by a return to expected GFR values in the Spring of 2021.

Additionally, the sizable 7.7 and 8.2% population-level reductions in birth counts in December 2020-January 2021 could partially be a mechanical artifact of closure of Assisted Reproductive Technology (ART) clinics in March and April of 2020. In California, about 2.4% of births or 10,000–11,000 births each year are supported by ART (Sunderam et al. 2022). To assess the role of access to ART, we create a time series on births that were not supported by ART. These estimates are plotted in open red circles in Figure 2. We find that ART closures contributed marginally to the December 2020 and January 2021 estimates; non-ART births fall by a similar magnitude as all births below expected values in both months. Because ART procedures can take many months to result in a pregnancy, ART clinic closures in Spring 2020 could contribute to births many months into 2021, however such effects are not detectable in the estimates shown in Figure 2.

Population Heterogeneity:

Aggregate population trends could mask substantial heterogeneity driven by migratory flows, biological, or behavioral adaptations. To address sources and patterns of heterogeneity, we consider multiple forms of variation across population subgroups in the estimates.

Given restrictions to international migration triggered by the pandemic, international migration could have a substantial impact on population change. Our analysis of the GFR points to a weaker decline in fertility in December 2020-January 2021 than obtained from birth counts, suggesting a reduction in the number of births to foreign-born people. We start, therefore, by examining differences by nativity. Figure 3 plots ARIMA estimates for foreign-born and native-born birthing people. The reduction in birth counts in the Fall of 2020 and Winter of 2021 is much larger for foreign-born than US-born birthing people, reaching between 10 and 15 percent between August of 2020 and February 2021 among the foreign-born population. This finding is consistent with increased international out-migration and reduced in-migration driven by long-term travel restrictions and slowdown of visa applications. The number of births to foreign-born birthing persons returned to its counterfactual level based on pre-pandemic trends only towards the end of 2021 and remained aligned with expectations thereafter.

Figure 3. Proportional deviation in birth counts from expected values by birthing persons’ nativity, March 2020 - November 2022.

Note: Estimates from ARIMA models analyzing monthly birth count values divided by the mean pre-pandemic (January 2014 – February 2020) monthly birth count. Estimates indicate the proportional deviation of birth counts in each month relative to the predicted value for the month based on trend, seasonal patterns, and other sources of temporal autocorrelation observed pre-pandemic. ARIMA analysis estimated separately by subgroup. 95% confidence intervals shown.

This pattern does not explain the entirety of the population-level fertility reduction, however. Foreign-born people comprise about one-third of California’s birthing population. GFR estimates (Appendix Figure B3) show similar patterns, which could be interpreted as fertility reductions among remaining foreign-born residents alongside changes in net migration. At the same time, the quality of GFR estimates is conditional on the ability of the CPS data to capture state-level short-term fluctuation in the foreign-born population. We return to this issue in the final section of the study.

Patterns are different for the native-born group: We see a much smaller reduction in the Winter of 2020/2021, reaching 3.8% below-expected birth counts in January 2021. Starting in Spring of 2021 we observe a sustained increase in births to US-born mothers, which remain above expectation until the Spring of 2022. The highest increase in birth counts occurs in the Winter of 2022, affecting conceptions occurring approximately in Spring 2021, a period characterized by a sharp decline in COVID infections after the surge driven by the Delta variant in the winter of 2020–2021 (Torche and Nobles 2022). Fertility increase among U.S.-born birthing people is larger and more persistent than the prior decline in Winter of 2020/21 and suggests a remarkable departure from the long-term fertility decline since the late 2000s. Analysis of population heterogeneity by age, parity, and socioeconomic status among US-born birthing persons further elucidates the factors driving this increase in birth counts in late 2021 and early 2022.

Figure 4 plots estimates from ARIMA models fitted separately by the age of the person giving birth (panel a) and by parity, or birth order (panel b) among U.S.-born birthing persons. We find evidence of substantial divergence by age. Early-pandemic birth counts fell most dramatically for people age 35 and above, reaching 10% below expected values in November 2020, 15% in December, 13% in January 2021 and 7% in February before returning to expected values in March 2021. No birth “rebound” is observed among this population later in 2021 or during 2022. The increase in birth counts since the Spring of 2021 is driven by birthing persons younger than 25 years old; birth counts increased substantially among this group, with persistently higher counts by around 10% from the Spring of 2021 to the end of 2022. Less variation in estimates emerges across birth parity. Declines in birth counts by 5–10% below expected values in Winter 2020/2021, consistent with fertility delays, are concentrated among multiparous women. In contrast, among primiparous women, birth counts did not decline by much in the early stage of the pandemic.

Figure 4. Proportional deviation in birth counts from expected values by birthing persons’ age and parity, U.S. born persons, March 2020 – November 2022.

Note: Estimates from ARIMA models analyzing monthly birth count values divided by the mean pre-pandemic (January 2014 – February 2020) monthly birth count. Estimates indicate the proportional deviation of birth counts in each month relative to the predicted value for the month based on trend, seasonal patterns, and other sources of temporal autocorrelation observed pre-pandemic. ARIMA analysis estimated separately by the age of the person giving birth (panel a) and by parity, or birth order (panel b). 95% confidence intervals shown.

Combined, trends by birthing persons’ age and parity suggest a substantial short-term decline in births as an immediate response to the pandemic without a later-pandemic rebound among older and multiparous women – a potential quantum shift that could result in a small reduction in cohort completed fertility for older birthing people.

The notable increase in birth counts among younger and primiparous birthing persons starting in the Spring of 2021 continues into 2022 and shows no evidence of abating, suggesting a slowdown in the decade-long pattern of fertility delay in the U.S. for these populations. In the absence of information on fertility intentions, it is difficult to interpret these births as a behavioral response to changing economic conditions–versus increased difficulty achieving fertility avoidance. What is clear is that the fertility increases observed into 2022 do not involve the same population in which we see large pandemic-associated declines earlier in the pandemic. In other words, these different trends may recoup fertility at the population level, but some families who delayed fertility in response to the pandemic’s onset are likely to reach their mid-forties with lower completed fertility than might have occurred in the absence of the pandemic.

Differences by educational attainment are also detectable. The most notable source of variation is in the size of the short-term decline in birth counts in the Winter of 2021. Among people with graduate education, birth counts were 9% lower than expected in December 2020 but returned to expected values in January 2021 (Figure 5, panel a). Reductions among those without a college degree in Winter of 2021 were smaller (and mostly not significantly different from zero at a conventional 95% confidence level). Counts are elevated for most of 2021 among college-educated people, and people with a graduate degree, indicating a marked slow-down of the post-Recession fertility decline in 2021 for highly-educated native-born people in California. For people with a graduate degree, this continues into early 2022 and though not precisely estimated, remains substantively large for most months of 2022.

Figure 5. Proportional deviation in birth counts from expected values by birthing persons’ education and residential socioeconomic resources, U.S. born persons, March 2020 – November 2022.

Note: Estimates from ARIMA models analyzing monthly birth count values divided by the mean pre-pandemic (January 2014 – February 2020) monthly birth count. Estimates indicate the proportional deviation of birth counts in each month relative to the predicted value for the month based on trend, seasonal patterns, and other sources of temporal autocorrelation observed pre-pandemic. ARIMA analysis estimated separately by education (panel a) and zip code socioeconomic status index quartile (panel b). 95% confidence intervals shown.

The early-pandemic decline in birth counts among more advantaged birthing persons is observed to a lesser degree when socioeconomic status is measured by resources of the zipcode areas in which birthing persons live. Reductions in the Winter of 2020–2021 can be detected across communities at all levels of economic advantages, though the drop in birth counts was somewhat more pronounced and persistent in the more resourced communities (Figure 5, panel b).

Finally, figure 6 plots differences by the race and ethnicity of the person giving birth (panel a) among the U.S. born. We find noticeable differences across groups: small fertility reductions in December 2020 – February 2021 among Hispanic and white populations, large reductions in Asian populations, and no deviation from expected values among Black populations. Notably, the late-pandemic increase in birth counts since the Spring of 2021 is largely driven by the Hispanic population. In contrast, declines in birth counts are observed among US-born Blacks and Asians during 2022 (with the caveat that estimates for these groups contain high uncertainty given the combination of forecasting uncertainty and smaller sample sizes).

Figure 6. Proportional deviation in birth counts from expected values by birthing persons’ race, ethnicity, and socioeconomic advantage of community of residence, U.S. born persons, March 2020 - November 2022.

Note: Estimates from ARIMA models analyzing monthly birth count values divided by the mean pre-pandemic (January 2014 – February 2020) monthly birth count. Estimates indicate the proportional deviation of birth counts in each month relative to the predicted value for the month based on trend, seasonal patterns, and other sources of temporal autocorrelation observed pre-pandemic. ARIMA analysis estimated separately by subgroup. 95% confidence intervals shown.

These group differences partially mask what appears to be a socioeconomically-graded phenomenon. When we examine fertility change among birthing people living in the highest SES communities (those in the top SES quartile), we observe pandemic-associated fertility reductions across all racial and ethnic groups ranging between 5% and 20% in the last months of 2020. Although the timing and magnitude of the declines vary across groups, in all cases the declines are short-lived and in all groups experience recovery in birth counts starting in early 2021 (Figure 6b).

Though the reduction of births supported by assisted reproductive technology is clustered among older, higher SES individuals, ART closures do not account for the disparate drops in fertility among these populations. Figures C1 and C2 in the Appendix plot ARIMA estimates for births conceived without ART by age and education. Among the oldest and highest educated people, non-ART births fall by proportions similar in magnitude to that observed for all births among these groups.

DISCUSSION

Our findings show that the widely expected “pandemic baby bust” did not materialize in California. We found a discrete, short-term decline in the number of births in Winter 2020/2021 largely driven by births that would have been conceived in March and April 2020, in the context of the declaration of COVID-19 as a global pandemic and the implementation of stay-at-home policies. By March 2021, the number of births returned to levels consistent with a pre-pandemic trend for most population subgroups examined in this analysis. If anything, we detect a stall in the longer-run fertility decline trend during late 2021 and early 2022.

The most novel set of findings emerges from our analysis of heterogeneity in fertility responses to the COVID crisis. We find that births to foreign-born people fell markedly from May 2020 to November 2021, largely driven by changes in net migration rather than fertility behavior. Among U.S. born people, he discrete decline in the number of births in the winter of 2021 was driven by birthing persons who were older, had higher levels of schooling– particularly those with graduate degrees– and who lived in socioeconomically advantaged communities. We also detected some variation by race and ethnicity, especially a stronger decline in the number of births among Asian birthing persons, but the decline in fertility among highly-educated groups occurs across all racial/ethnic groups. After this decline, there is some indication of an increase in the number of births beyond trend-based expectation beginning in Fall 2021 and maintained through November 2022, in particular for persons who are younger, primiparous, highly-educated, and Hispanic.

These findings align with results from recent research in the United States and other wealthy countries indicating that pandemic fertility reductions were likely much smaller and shorter lived than initially predicted. The findings are somewhat surprising in light of the significant economic shock accompanying the pandemic in the U.S., and the theoretical expectation that some people would delay fertility until economic conditions improved and uncertainty declined. Indeed, our findings do not support the hypothesized stronger fertility response among disadvantaged populations more affected by the acute economic shock early in the pandemic followed by a robust economic recovery in California. The short-term fertility declines observed are concentrated among the most resourced people, who were least exposed to the pandemic’s unemployment shock and the accompanying state and federal economic relief efforts. In this sense, the present study adds to a growing line of work suggesting that the predictive power of macroeconomic indicators for U.S. fertility change may be weaker in the contemporary period (Buckles, Guldi, and Schmidt 2022; Kearney et al. 2022; Manning et al. 2022; Seltzer 2019). This evidence presents an opportunity for social scientists to grapple with other, difficult to measure, sources of long run reductions in population fertility beyond macroeconomic indicators, such as subjective perceptions of risk, expectations, and aspirations (Guzzo 2022).

Noticeably, the sharper decline in number of births among women 35 years old or older was not accompanied by a commensurable rebound in the summer of 2021. Given age-related fecundity constraints, this pattern could translate into forgone fertility among this age group. Even if this cohort effect were not sufficient to alter population-level trends, the pattern has consequences on later-life health and wellbeing for the cohort that may have seen their fertility intentions permanently altered.

The socioeconomic stratification in fertility decline suggests that the reduction in number of births in the early stage of the pandemic was likely a purposive response, potentially based on heightened perceived risk and uncertainty even among segments of the population that did not bore the brunt of the health and economic shock. This finding is consistent with the insight that fertility might be more sensitive to change in short-run expectations than previously thought (Buckles, Hungerman, and Lugauer 2021). The finding that births achieved by assisted reproductive technology (ART) contribute to a very small portion of this decline is consistent with an interpretation that focuses on intentional behavioral responses, given that the decline in access to ART was an external constraint beyond the control of populations intending fertility.

The socioeconomic differences observed here are also consistent with a model of fertility change in which many people desired to delay births and people with more socioeconomic resources faced fewer barriers to doing so. A national study of desired fertility in April/May of 2020 reveals that 34 percent of women sought to delay childbearing to have fewer children as a result of the pandemic, and that this change in fertility preferences was stronger among socioeconomically disadvantaged women (Lindberg et al. 2020). The discrepancy between desired and achieved reductions in fertility is almost certainly related to a socioeconomic gradient in the ability to actualize preferences (Bailey, Bart, et al. 2022). As is well established, people with lower incomes, living in less advantaged communities face substantially larger barriers to access contraceptives; multiple studies indicate that this pattern persisted or was exacerbated in the context of the pandemic (Diamond-Smith et al. 2021; Lin et al. 2021; Smith et al. 2022).

Indeed, a key limitation of research using vital statistics records is the inability to distinguish intended and unintended fertility. This is a tradeoff incurred by the choice to use administrative records, which allow us to study fertility among the socioeconomically most disadvantaged people, who are underrepresented in survey research. Future studies may make headway leveraging survey data attached to administrative records to achieve partial identification (i.e. estimate bounding) of shifts in intended and unintended fertility, or use creative approaches that estimate effects from measures of contraceptive efficacy (Bailey, Bart, and Lang 2022).

A second limitation of this study is that it is entirely silent on partnership and marital fertility. The California Department of Public Health is unusual in that it does not release marital status of the person giving birth in the birth records. Research with early-release data from other state departments of health may be able to fill this important gap.

A third limitation of using birth records to study fertility is that we are only able to capture live births. A behavioral interpretation of the patterns here ignores the possibility of changes in miscarriage and abortion. Studies to-date do not indicate increases in abortion in California in March-May 2020 that could plausibly account for late 2020 fertility reductions (Jones, Kirstein, and Philbin 2022). Fetal death is too rare to contribute to the size of changes demonstrated here. It is possible that early miscarriage rates increased in Spring 2020. We cannot detect that with these data and it remains an outstanding question for future work.

A final limitation of this study is an issue faced by all population scientists conducting research in the U.S. The difficulty of measuring short term population fluctuations is well-known. The Current Population Survey (CPS) is probably the best resource for social scientists studying the U.S. as it provides monthly information about population composition. However, this data source is not without measurement error. The estimates of GFR values using the CPS for the period of study yield results that for the most part are consistent with birth counts; fertility at the youngest ages is a notable deviation. Nevertheless, it remains likely that changes in population sizes contributed to the reductions in birth counts in the Fall of 2020 in ways that are imperfectly captured in the CPS data. This issue is exacerbated for smaller groups of the California population, such as U.S.-born African Americans, graduate-degree holders, and people giving birth at the youngest ages.

The findings have an additional, general implication for the study of welfare in the context of the ongoing pandemic, and of large-scale environmental shocks more generally. Even small and short-term heterogeneous fertility responses to the pandemic have mechanical consequences for the composition of birth cohorts. The observed concentration of a fertility decline among the most advantaged groups will result, ceteris paribus, in a cohort with more limited family resources and that would likely benefit more from public support. This compositional change, even if brief, is potentially consequential for the cohorts born during the early stages of the pandemic as they enter the educational system, the labor market, and other institutions–and is therefore relevant for future evaluation of the medium- and long-term impacts of the pandemic on child and family welfare. In a relevant example, Beach et al.’s (2022) study demonstrates how the 1918 flu pandemic changed the composition of birth cohorts, making incorrect earlier conclusions about the long-run impact of that pandemic on gestating cohorts (Almond 2006).

Looking ahead, this study, like many others describing fertility changes around the world in the context of the pandemic, relies on innovation in the ability to access near real-time administrative data (e.g., state-level vital statistics, electronic health records–Bailey, Currie, et al. 2022; Rangel et al. 2020), which are a powerful resource to learn about changing reproductive behaviors. Interpretation of patterns in these data for population welfare are significantly hindered by the absence of information on fertility intentions. Social scientists would benefit from similar innovation in the measurement of intentions or constraints on achieving intended fertility–e.g., records of contraceptive requests, fulfilled prescriptions, contraceptive prices, reproductive health care waiting times, among many other data sources, that may be an avenue to shed light on subpopulations most in need of support concerning unplanned fertility in the context of large-scale social and economic change.

Data Availability

Restricted access data can be obtained from the California Department of Public Health (https://www.cdph.ca.gov/). Code used to generate estimates in this study are accessible on the first author’s GitHub page.