The US baby boom and the 1935 Social Security Act

ABSTRACT In 1935, the United States passed Social Security Act (SSA) providing financial security to American families. I use the individual census data for 1940 and 1960 to show that women from states that allowed for more social spending under the SSA had substantially more children than women from states that allowed for lower social benefits. I also use a new panel of state-level fertility by parity between 1935 and 1959 to show that family allowances were connected to first, second and third parities, but that there was a differential effect according to the different social programs and race.


Introduction
In the United States, the fertility transition (the process by which the country went from high to low fertility) took place in the nineteenth century. Across the nineteenth century, white women could expect seven or more children and African American women eight. By contrast, today many couples are just replacing themselves (or just below it). This longrun decline was interrupted between 1935 and 1959, when the number of births started to increase. This swing came to be called the baby boom and was followed by a baby bust where, by 1960, fertility resembled its historical trend. The sheer size of the baby boom generation (some 75-80 million) signaled a turning point in American history and magnified its impact on society. For instance, it prompted a building boom in housing, a shift in labor supply, and as baby boomers grew and prospered in the 1980s, their buying habits determined the course of many consumer industries. In most recent decades, the baby boomers have also confronted the challenge of population aging, threatening longstanding formulas on pensions and the provision of healthcare.
While there is a growing literature documenting how unusual was the fertility transition in the United States (i.e. the decline came very early in the nation's development and in the European mirror, where only France had a comparably early onset) the causes behind the American baby boom, the defining episode of the twentieth century, remain unclear (Guinnane, 2011). For instance, Albanesi and Olivetti (2014, p. 226) stressed that 'despite the remarkable magnitude of these fluctuations in fertility and their clear economic and social relevance, their origins are still poorly understood.' In a similar vein, Bailey and Hershbein (2018) pointed out that 'more theoretical and empirical research on the explanations for baby boom and bust is needed to vet existing theories and, perhaps, to develop new explanations. ' The war mobilization hypothesis -when after World War 2 (WW2) around 16 million young men returned home and resumed work and family life -is perhaps the one that has received the most attention (Acemoglu et al., 2004). However, this 'tempo' hypothesis could not explain why the US baby boom took off in the mid-1930s. Recently, Brodeur andKattan (2022) have also shown that higher mobilization rates led to lower fertility. A related explanation is the entry of women into the labor market during WW2 and their departure at the end of the war (Doepke et al., 2015). Yet, Acemoglu et al. (2004) noted that the effect of wartime female participation on wages dried up by 1950. Goldin (1991, p. 775) also commented that the impact of the war affected women of different ages, and not just young mothers: 'it seems unlikely, though, that the impact of the war worked entirely through young women who were the mothers of the "baby boom" in the 1950's. ' Albanesi and Olivetti (2014, p. 229) also said that 'this explanation is inconsistent with the fact that fertility began to rise before the war and with the limited direct impact of wartime female participation on labor market conditions.' Greenwood et al. (2005) also noted that the baby boom was not just the result of couples postponing childbearing after WW2, as the number of lifetime births per woman increased as well. See also Bailey and Hershbein (2018) and Greenwood et al. (2005).
Easterlin's hypothesis of 'relative income'-whereby favorable labor market conditions increased the desired fertility -has also been debated (Easterlin, 1976(Easterlin, , 1987(Easterlin, , 2000. Bailey and Hershbein (2018) and Haines (2018) noted that despite major economic fluctuations after the 1960s, fertility did not continue to cycle. Bailey and Collins (2011) also criticized Greenwood et al. (2005) view of the baby boom as a technological shock, as they found that the Amish community also experienced it (see also Lewis, 2018). Other explanations account for the relative size of female wages compared to men (Galor & Weil, 1996), cultural changes (Lesthaeghe & Surkyn, 1988), rural migration and the agricultural revolution (Baudin & Stelter, 2018).
In this paper, I explore the impact of the first major welfare reform in US history on the baby boom story: the 1935 Social Security Act (SSA). I use the 1% individual data from the decennial census to explore spatially the impact and the potential mechanism of the SSA on the baby boom by different maternal age groups (Ruggles et al., 2021). Fixed effects models, interaction terms, an instrumental variable strategy and placebo tests on nonmaternal funding help to isolate the effects of SSA on fertility. To shed further light on the mechanisms, I also describe the effect of SSA on fertility using a state-level panel by parity and maternal race, with state and year fixed effects using annual data from 1935 to 1959.
Overall, I find a consistent impact of social plans on fertility for mothers of different ages. Yet, the effect is mostly concentrated in mothers aged between 20 and 34 years. Here, each one-standard-deviation increase in new social spending was associated with between one-tenth and one-third of one standard deviation of the dependent variable (fertility rate). The size of the impact varies depending on how social aid is measured. The new social scheme connected with white-and non-white mothers and not only encouraged first-time mothers but helped to create an environment of financial security to increase family size.
The idea that the SSA helped to raise US fertility has already been suggested by previous scholars. For instance, Albanesi and Olivetti (2014, p. 235) noted that 'the Social Security Act spread the benefits of health insurance more broadly for maternal conditions ' and Fishback et al. (2007, p. 12) that 'a more substantial social safety net contributed significantly to the leveling off of the fertility rate in the late 1930s.' However, Fishback et al. (2007) paper stops in 1940, without contributing further to the baby boom literature. There is also empirical evidence that early Mothers' Pensions positively correlated with fertility (Moehling, 2007). Settersten et al. (2021) also discuss the financial decisions around delayed fertility during the Depression in the context of San Francisco.
The paper also connects with Lindert (1978), who linked the baby boom to a set of new fiscal reforms on taxation that occurred under the SSA, with fiscal benefits to incentivize earlier marriage and higher fertility. In addition to this, the link between the SSA and the baby boom also needs to be seen in the social and cultural contexts of the 1930s, where Roosevelt's social reforms sought to encourage family values with a popular culture that celebrated pregnancy, parenthood, and large families. Outside the US, Boyer (1989) found evidence that the social policies of the Poor Laws in England also shifted fertility up and Buttner and Lutz (1990) that after WW2, the new social policies implemented in the German Democratic Republic also raised fertility (while declining in the Federal Republic).
In more modern settings, scholars also find that higher governmental support for families in cash benefits and family allowances positively connect with fertility (Brewer et al., 2012;Castles, 2003;Gauthier & Hatzius, 1997;Kalwij, 2010;Milligan, 2005;Moffitt, 1998;Schellekens, 2009;Vos, 2009). For instance, in a cross-national study of 16 Western European countries, Kalwij (2010, p. 503) concluded that 'the results show that increased expenditure on family policy programs . . . generates positive fertility responses.' There is also evidence that tax incentives increase fertility (Whittington et al., 1990), as do subsidies for child-care costs (Blau & Robins, 1989), including maternity benefits such as maternity leave (Averett & Whittingtont, 2001). Evidence for the Aid to Dependent Children (ADC), a program of the SSA, also positively connected with fertility (Powers, 1995) and modern US social security programs such as Aid to Families with Dependent Children (AFDC) also show a positive link with fertility and childbearing decisions (An et al., 1993;Argys et al., 2000;Camasso, 2004;Grogger & Bronars, 2001;Powers, 1995;Robins & Fronstin, 1996), with evidence that cap policies can mitigate the effect of such programs (Fairlie & London, 1997). 1 The paper continues as follows. The next section provides more details on the welfare reforms under the SSA. Sections 3 and 4 are the main empirical sections exploring the association between the SSA and fertility using individual-and state-level data. Section 5 concludes.

The social security act
The SSA represented a fundamental change in the way relief was provided to American families. Before the SSA, support was seen as the responsibility of families and local administrations. It was mainly driven by private and philanthropic organizations such as Poor Houses and churches. It was a patchwork system that was inadequate for a massive catastrophe, like the Great Depression. When private assistance dried up early in the Depression era, Roosevelt used federal funds administered by states and localities to provide relief to vulnerable people. Yet, temporary emergency programs were calling for something permanent, and a new landscape in the way aid was provided to vulnerable families, pushing states and the federal government into a new role (Fishback, 2017(Fishback, , 2020. Hence, social security came in the summer of 1935, with the launch of the SSA. The SSA developed a new general welfare scheme of social security and an insurance against unemployment. It targeted particularly vulnerable groups, such as the elderly, unemployed, dependent children, maternal and child welfare and public health work. The Social Security Board was given responsibility for the federal administration of most of the grants-in-aid features of the SSA, except for maternal and child-health services (administered by the Children's Bureau) and public health work (administered by the Public Health Service). As noted by Fishback (2020, p. 321) 'State governments determined benefit levels in many public assistance and social insurance programs.' Since the level of benefit was set by states, generosity greatly varied across states and time. 2 There was also heterogeneity across programs but, most of the times, the SSA introduced federal matching grants to the states, and states continued to raise benefits relative to the poverty line.
Title V of the SSA was designed to care for mothers and children, providing maternal and child-health services and child-welfare services. Hence, funds were spent on obstetric provisions such as prenatal care, baby clinics, immunization and public health nursing, as well as medical coverage and hospital services Lesser, 1985;Oettinger, 1960;Schmidt, 1973). As Schmidt (1973, p. 423) explains, 'at first, the Maternal and Child Health programs were largely devoted to such services as pre-and postnatal clinics and child health clinics and to training of professional personnel. By 1937, however, the Bureau's advisory committee on maternal welfare recommended that the program be enlarged to provide medical and hospital care of mothers during labor and delivery.' Among many services, states were also mandated to make diagnostic services available free of charge without the requirement of economic status or legal residence (Albanesi & Olivetti, 2014).
From its inception, the SSA continued to expand. Between March 1943 and 1946, it launched the Emergency Maternity and Infancy Care (EMIC) for wives and children of US servicemen deployed during WW2. The EMIC included medical, nursing and hospital care for the prenatal periods, delivery and six-week postpartum, with an estimated number of 1.5 million mothers receiving care. During WW2, the SSA also implemented additional programs due to the withdrawal of doctors and nurses who joined the Armed Forces (Lesser, 1985;Oettinger, 1960;Schmidt, 1973), and in 1946 also launched the Hospital Survey and Construction (Hill-Burton) Act, to create sufficient beds to attend to at least 4.5 per 1,000 of the US population in hospitals. This Act also enforced no discrimination on racial grounds, and in 1975 was amended to become the Public Health Service Act.

Social security act effects on fertility
Since the SSA was not uniform across states, I start exploiting the aggregate state-level changes in the new social plans. As benchmark, Equation (1) investigates if the foundation of the social security affected the baby boom. That is, if states that allowed for more early benefits in social security between 1940 and 1960 experienced a larger baby boom. Thus, I rely on the assumption that spending in the early years (i.e. 1936 and 1944) is a sufficient statistic for state commitment to social welfare programs. Specifically, pooling the data from the census of 1940 and 1960 I estimate the following specification: where the dependent variable y is the measure of fertility, either calculated as the number of own children under the age of five living in the same household or the number of children ever born for the women i in the state s and year t, with the model displaying results for mothers of different ages, as defined by a. Here, I use the 1% Integrated Public Use Microdata Series (IPUMS) of the decennial censuses (Ruggles et al., 2021). SSA s is the state spending under the SSA, measured as the average payment per capita between 1936 and 1944 using data from the Social Security Bulletins (several issues). If instead of 1944 I tested for alternative periods I get the same overall findings (unreported here). Additionally, Λ´X ist is a vector of individual-level controls available in the IPUMS data, μ s is a state fixed effect and D 1960 is a dummy variable for 1960. Results exclude women living in group quarters and Alaska and Hawaii as until the 1950s they did not have statehood. Descriptive statistics of all variables throughout the paper are available in Table A1.
This design is very similar to those of Doepke et al. (2015, 1,040, equation (1)) and Acemoglu et al. (2004, 521, equation 8), where the key parameter of interest β is the interaction of the social security spending with the 1960 dummy, meaning that if β > 0, fertility would have increased by more between 1940 and 1960 in states with higher early social security spending. All regressions use the PERWT sample weights and include dummy variables for observational data on year, age, race and state of birth. Women are assigned to states based on their state of birth and standard errors are clustered by the year and state of birth. I use the state of birth, instead of the state of residence, as many programs of the SSA had residence requirements. Hence, I explore whether women growing up in a more generous welfare regime and that greater financial support were induced to have more children. For models using children under the age of 5, I include the number of children younger than 5 and for children ever born, parity is restricted to the seventh birth to avoid problems of potential outliers and measurement error. Sample size when fertility is measured as the number of children ever born is substantially reduced. This is because this variable was only available for a small sample line group of women in 1940. Finally, e ist is the error term. Table 1 displays the main results with each entry showing the estimate of the interaction term β for a different specification. Results for fertility show that women from states with more generous SSA plans but who are otherwise of similar characteristics had substantially more children when compared to women born in states with less generous social plans under the SSA. Most results for fertility for mothers aged between 20 and 34 years are statistically significant, but given the magnitudes at play (i.e. number of children under the age of 5 or children ever born) the overall individual effect is small. The point estimate implies that a 10-percentage point higher per capita social security plans translated into 0.006 more children under the age of 5 in 1960. For children ever born, 10percentage point higher per capita social security plans translated into 0.014 more children. For comparison, Doepke et al. (2015) estimated that at a 10-percentage point increase in the mobilization rate increased the number of children under 5 years of age by 0.14 and the number of children ever born by 0.20. When I add controls for years of education and farm status, the size of the coefficient is reduced, but signs and levels of significance are about the same. The same is true when adding the marital status dummies (same signs, levels of significance but magnitudes are lowered). By maternal age group, most of the effect is concentrated in mothers between 25 and 29 years and for mothers aged 20 and 24 years and 30 and 34 years, when fecundity is still at its peak. Since I assigned women to their states of birth, women who were in the 25-29 group in 1960 were infants and children in 1936, and many of them might have received benefits directly throughout their life and greater financial support at the time of pregnancy, or indirectly through their parents or relatives. These women may be differentially impacted compared to the previous cohorts, as were the first cohort born within a system of social security, and they were growing as social welfare expanded. The effect for mothers after they become 36 is reduced, losing its value as predictors as results are no longer statistically significant. This is probably because women closer to the end of their fecund period (40-45 years old) in 1960 had much of their childhood pre-welfare. These results are in line with the mean age at first birth, which was around 20-25 years in 1935.
Since the SSA was a compendium of different social plans, it is also possible to look at the channels of this association. In Table A2, instead of using total SSA spending, I use spending under the main individual programs of the SSA (the per capita spending in the main programs of the SSA) and fertility as measured as having children under the age of 5. I find that most of the impact is driven by spending in Maternal and Child Health services and as noted in the introduction, this was a program targeted at improving maternal conditions and providing security for mothers to deliver safely. In the 1940s, this program was also greatly expanded, which has the potential to explain not only why fertility started to increase in the mid-1930s but in the post-WW2 baby boom as well, with the EMIC program and the Hospital Survey and Construction Act. Hence, when controlling for Maternal and Child Health Services instead of all SSA spending, the size of the impact of fertility (in terms of children under 5 years of age) increased. Point estimates are very similar for ADC, a program designed to take care of children and maternal conditions in poor households. This program was reformed in 1962 and renamed as AFDC, with evidence that the AFDC positively correlated with fertility (An et al., 1993;Argys et al., 2000;Camasso, 2004;Grogger & Bronars, 2001;Powers, 1995;Robins & Fronstin, 1996).
A similar kind of effect holds for unemployment and general relief in order to help vulnerable families. The effect of the lion's share of the SSA, the Old-Age assistance, a system of pensions for the elderly (Galofré-Vilà et al., 2022), shows the smallest coefficients from all programs, and despite being statistically significant, its impact given its small coefficient seems highly limited. Finally, as a placebo test, it is also reassuring that the impact of spending under Aid to the blind, which is the only program of the SSA that I would not expect to affect fertility (as it targets a very particular group of the American population) is uncorrelated with fertility, displaying very low p-values.
In Table 2 I also show a more demanding specification in order to deal with problems of reverse causality and omitted variable bias. For instance, it is possible that SSA spending was not randomly assigned, that farming and race connected with fertility in ways unrelated to SSA or that there were also differences in general economic trends between agricultural states affecting levels of fertility. Since two basic concerns are the agricultural states and racial inequalities, I partly address such concerns by including measures of concepts that may affect fertility listed by the Social Security Board that affected SSA in the main regressions (Social Security Bulletin December 1965). Specifically, as in Doepke et al. (2015), I add the fraction of African-American males and the fraction of men aged 13-44 with farm occupations (all in 1940) interacted with the 1960 indicator variable and with the same controls as in Table 1. Results for fertility show that these controls are important, as adding a control for the black share (column 2) reduces the size of β and also levels of significance, but results stay significant at the 10% level. However, controlling for the share of farming (column 3) has only a limited effect where the size of β remains about the same and potentially correcting some of the problems of omitted variable bias.

Panel data with fixed effects
While individual-level data avoid potential problems of ecological fallacy, so that higher levels of aggregation do not correspond to individual-level causation, I next use a novel state-level panel data for 1936 to 1959 to explore with annual data the kind of mothers that could be affected by the new social scheme. Specifically, I am now testing for a contemporaneous connection between SSA spending and fertility rates. Since I added state fixed effects, a positive effect of SSA spending needs to be interpreted as an increase  The outcome variable is the total fertility rate (births adjusted by the female population between 14 and 44 years and per thousand) in 4 parity groups. I compute fertility rates according to different parities for all, white and non-white mothers. The method of estimation is OLS. All figures are adjusted for state-and time -varying controls as defined in equation 2. All models add state and year fixed effects with robust standard errors clustered at the state level. For details on data see text, * p<0.1, ** p<0.05, *** p<0.001 in SSA spending leading to a contemporaneous increase in births. I also use this novel state-level data to quantify the effect of social security on the overall baby boom. I begin exploring the impact of the new social aid of fertility by maternal age with the following equation: where TFR is the total fertility rate for mothers who gave birth in the state s, year t (t = 1936, . . . , 1959), according to 5 parity groups defined by a (all births, 1st child, . . . ., 4th child) and maternal race r (all mothers, whites and non-whites). The number of births is based on a novel data set using data from the Vital Statistics of the United States, and age adjusted population data from Haines (2010) are linearly interpolated between census years.
In Table 3, I display the results using data in the per capita spending under the SSA (the first four columns) and Maternal and Child Health services (in subsequent columns) using data from the monthly bulletins of the Social Security Board. X st is a vector of controls for state-and time-varying controls reflecting factors influencing the baby boom, such as personal income per capita (data are from the Bureau of Economic Analysis), education (the number of pupils in primary and secondary schools adjusted by the number of teachers), the urban share and the number of marriages per thousand population (these last controls use data from the Statistical Abstracts from the United States). The models also include state (µ s ) and yearly (δ t ) fixed effects, allowing for robust standard errors clustered at the state-level and with e st being the error term. Here, I also standardize data to have a mean of zero and a standard deviation of one, so coefficients across models are directly comparable.
Results show that for second parities, a one-standard-deviation increase in payments per capita under the SSA is associated with an increase in the fertility rate (in standard deviation terms) of 0.007. The size of the impact greatly increases when one looks at Maternal and Child Health services. Here, also for second parities, one-standard-deviation increase in payments per capita under Maternal and Child Health services is associated with an increase in the fertility rate (in standard deviation terms) of 0.359.
By race, results considering all per capita SSA spending show that most social aid is connected with white-and non-white mothers. Although at the beginning SSA programs were not specifically inclusive of African Americans (Fetter, 2017;Galofré-Vilà et al., 2022;Galofré-Vilà, 2020), the SSA was also not directly exclusive of them (Fishback et al., 2007), and as the program expanded in the 1940s and 1960s, more African American women benefited from the new social scheme. The fact that the new social programs comparatively affected more non-white mothers (i.e. larger size of coefficients) can be explained because of their greater vulnerabilities and economic insecurities. This finding can be also related to the overrepresentation of non-whites on public relief rolls. 3 Evans (1986) pointed out that 'it is very clear that increases in fertility rates at the higher parities were a much more important part of the baby boom among nonwhites than among whites.' Social benefits in columns 1-4 mostly connected with later parities, with SSA benefits affecting second and third parties. While this is consistent with the idea that completed fertility of white mothers is close to 3 in the 1900s and rises to a peak of 3.5 in the late 1930s (Evans, 1986), 4 this finding suggests that the new social scheme not only encouraged first-time mothers, through maternal aid (Albanesi & Olivetti, 2014), but helped to create financial security to increase family size. Bailey and Hershbein (2018, p. 85) also note that 'increases in the period fertility rates during the baby boom were achieved both by earlier and more universal marriage and by married women giving birth to more children. ' As already seen, the SSA not only included spending on maternal care but also considered vulnerable mothers through ADC or unemployment programs (Galofré-Vilà, 2023). Indeed, as I show in columns 5-8, when I only control for Maternal and Child Health services (instead of the overall SSA spending), results are mostly only statistically significant for first parities (column 5).
Several robustness checks bolster these findings. First, in Figure A1 I report the scatter plots for the unadjusted association between the total fertility rate and family allowances. Second, in Table A3 I show the results with logarithm transformations in all variables (i.e. using elasticities). Third, in Table A4 I show that results look similar without the main controls and only adding state and year fixed effects. This is because one threat to interpreting the findings is that average payments track state incomes quite strongly and those can change differentially across states over time. I also used lagged values of SSA variables to control for the time between conception and birth (unreported here). These results are robust with those presented in Table 3.
In Figure A2 I also drop one state at a time to show that the effects are not concentrated in a few individual areas of the country and that no individual state drives the estimates (outliers). In Table A5, I also split the sample by Southern and non-Southern states. Here, I find that the effects of the SSA are statistically significant in both cases for later parities, but that in Southern states, results display a negative sign for first parities. This is consistent with early work from Alston and Ferrie (1985) for young mother in the working age, who argued that southern legislators chose not to accept all the federal money to which they were entitled to maintain cheap labor and assure that those most needed in cotton cultivation were kept off the welfare rolls. Another worry is that crossstate differences in per capita SSA spending could be related to trends in fertility. In Table  A6, I have included a time trend to the baseline specification to show consistent results with those in Table 3. 5 In Table A7 I also show regressions that predict per capita SSA spending based on state-level economic and demographic conditions. While cultural characteristics such as education and marriages are not strong predictors of the SSA, income seems important. As noted in Section 2, under the budget-deficit principle, this is not a surprising finding and, in this case, income shows the expected negative sign. Here, income per capita is a seemingly proxy for 'richer' states and plausibly depicts the 'demo-economic paradox' (Simon, 1969). Yet, to explore that this is not driving the results, using the same motivation as in Fishback et al. (2007), 6 in Table A8, I instrumented the per capita SSA spending with the standard deviation of the presidential vote to the Democratic candidate from 1896 to the most recent election (see also Fleck, 1999). 7 Results in the 2SLS using this swing-voting measure are in line with the OLS, being just 15-20% larger (in absolute magnitude) than the OLS results, and suggesting that OLS results may not be highly biased.

Concluding remarks
Despite the renewed interest to investigate the causes of the baby boom, its origins are still unclear. After revisiting competing explanations, including the decline in maternal mortality, the entry of women into the labor market and the military mobilization during WW2, this paper adds an alternative account to the baby boom story: the role of the SSA. This mechanism complements and synthesizes the existing ones. For instance, the launch of the SSA was crucial for obstetric care and to reduce maternal mortality; it removed financial insecurities to American families; and took care of mothers whose husbands fought in WW2. Indeed, it is among the few explanations with the potential to explain why the baby boom started in the mid-1930.
Overall, I find a consistent impact of social plans on fertility for mothers of different ages. Yet, the effect is mostly concentrated in mothers aged between 20 and 34 years. Each one-standard-deviation increase in new social spending was associated with between one-tenth and one-third of one standard deviation of the fertility rate, the size of the impact varies depending on how social aid is measured. The new social scheme connected with white-and non-white mothers and not only encouraged first-time mothers but helped to create an environment of financial security to increase family size. Notes 1. Following Becker and Barro (1988), there is a literature arguing that pensions for the elderly, an important component of social security, tend to reduce (rather to increase) fertility in the rich world. 2. Public assistance programs ran on a budget-deficit principle, where needs and resources were evaluated, and the excess of needs over resources provided the basis for determining the payment. 3. For instance, using the language of the time, the Unemployment Relief Census of 1933 observes that 'there was almost twice as high a proportion of Negroes as of whites on relief for the United States as a whole.' See Galofré-Vilà (2023). 4. Data are from the Human Fertility Database. 5. A similar discussion on the inclusion of a time trend is made by Fishback et al. (2007, p. 7). 6. For instance, state-level policies could be correlated with levels of need/income, and levels of need might also affect the provision of welfare. 7. For this swing-voting measure I used the data from Fishback and Kachanovskaya (2015) from 1936 to 1941 and extended it using the official voting statistics for the period 1974-1959.