Introduction

1Total fertility in the United States fell from 7.0 in 1835, one of the highest rates in the Western world at the time, to 2.1 in 1935, one of the lowest (Coale & Zelnik, 1963; Hacker, 2003). In many respects, the United States is an ideal setting for studying fertility decline. The population was ethnically and racially diverse and experienced a wide variety of economic and environmental conditions, facilitating tests of hypotheses. Immigration to the US was episodic, with periods of German, British, and Irish dominance of the migrant stream from 1840 to 1890, followed by greater numbers of migrants from Southern and Eastern Europe after 1900. Immigrant groups came with different cultural ideas about fertility (King & Ruggles, 1990), but among most second generation groups – born and raised in America – fertility converged to native levels. Large racial, ethnic, and religious differences in fertility were complemented by geographic differences. These geographic differences were not merely a reflection of the spatial pattern of race and immigrant and second-generation settlement. Instead, they reflected spatial differentials in industrialisation, agriculture, urbanisation, school attendance, women’s labour force participation, population composition, religion, and occupational structure (Hacker, 2016).

2Although there has long been scholarly interest in the US fertility transition (Yasuba, 1962), researchers have been hampered by poor data. The diversity of the population and variation in state institutions contributed to substantial differences in the collection of natality data. Religious diversity means that scholars of the US fertility decline cannot rely on baptismal records for a nationally representative study. The national birth registration system which merely federated diverse state systems was not complete until 1933, near the end of the century or more of fertility decline. Few states had effective birth registration before the beginning of the twentieth century, well after such systems had been established in comparable countries (Higgs, 2004; Shapiro, 1950; Sköld, 2004). Samples of the pre-1940 censuses have helped address the lack of birth registration data. Until recently, however, sample densities were relatively low (1%), which has limited the ability to study contextual factors and smaller population subgroups.

3This paper leverages the analytical power of new IPUMS [1] complete-count microdata databases of the 1850, 1880, 1910 and 1930 decennial censuses (a joint on-going project between the Minnesota Population Center and Ancestry.com) to reexamine the US fertility transition (Ruggles, 2014; Ruggles et al. 2019). The combined dataset includes nearly 300 million individuals, spanning the beginning of the fertility decline in the mid-nineteenth century to within a few years of its temporary nadir in the mid-1930s. These data allow us to evaluate the behaviour of small population subgroups, including 18 different immigrant groups (identified by place of birth data collected by each census) and their second-generation descendants (identified by parental birthplace data beginning with the 1880 census). We include the free African American population in 1850, when free blacks constituted about 12 percent of the nation’s African American population, and all African Americans beginning with the 1880 census, 15 years after the abolition of slavery. As described in more detail below, the complete-count datasets also allow us to examine the potential impact of nearby patrilineal kin on couples’ fertility. We focus on whether the nearby presence of patrilineal grandmothers – the mothers-in-law of childbearing women – had an impact on couples’ fertility and whether the long-term decline in kin propinquity in the United States played a role in the fertility transition.

4Our study population includes all currently-married women age 20-49 enumerated by the 1850, 1880, 1910 and the 1930 censuses with spouses present in the same household. For this population we model recent net fertility (number of co-resident own children under age 5). This measure will understate actual fertility in the presence of infant and child mortality, which was high but declining in this era. Across the four census enumerations spanning eighty years, we use a consistent set of independent variables to evaluate the role of changing factors in the fertility transition and to decompose their contributions over time. We include measures often neglected by demographers, such as detailed nativity, kin availability, and a proxy of parental religiosity.

Prior studies

5Quantitative research on the US fertility transition has emphasised economic factors. In contrast to research on the European fertility transition (Coale & Watkins, 1986), cultural differences received relatively little attention in the foundational years of historical demography in the 1960s through 1980s. Because US fertility decline began when the nation was still overwhelmingly rural, researchers have focused their investigations on the possible role of changes in the agricultural economy on reproductive behaviour. Child-woman ratios are available at the county level between 1800 and 1860, from contemporaneous US census tabulations. For subsequent years they can now be calculated from complete datasets of the census, only recently made available (Ruggles, 2014). Differentials in child-woman ratios have been associated with differentials in the availability of land for farming, the price of local farms, and other measures of the agricultural economy. As land became scarcer or more expensive, parents adapted by limiting their fertility (Yasuba, 1962; Easterlin, 1976; Vinovskis, 1976; Easterlin, Alter and Condran, 1978; Smith 1987; Carter, Ransom and Sutch, 2004; Haines & Hacker, 2011).

6The introduction of a nominal, individual-level, census in 1850 and the development of microdata samples of these censuses since the 1970s has allowed greater focus on variations in individual fertility decisions in different contexts. Research on the post-1860 period, for example, has stressed the contributing roles of urbanisation, industrialisation, and compulsory schooling (Guest, 1981; Smith, 1996; Wanamaker, 2012). As with most cross-sectional studies of historical fertility decline elsewhere, however, causal pathways were likely complex and remain obscure. The negative relationship between education and fertility, for example – one of the most consistent findings in different societies and over time – might have been the result of many different causal mechanisms, including increasing parental costs associated with children’s changing roles from farm or industrial labourer to student (Caldwell, 1980; Guest & Tolnay, 1983). The increasing economic returns to education associated with industrialisation and occupational diversification likely resulted in shifting parental aspirations for their children, leading to greater investments in child “quality” through continued secondary and post-secondary education and lower investments in child “quantity” (Becker & Lewis, 1973). Increases in the education and literacy of parents themselves, and an associated increase in the supply of books, magazines and newspapers, resulted in greater exposure to secular culture and information on birth control and contraception (Brodie, 1994). Women’s education, in particular, resulted in women’s increasing power within the household vis-à-vis their spouses and increased the potential for participation in some occupations (teacher, nurses, etc.) in the paid labour force, both of which would have resulted in lower fertility. Research on the post-1860 period, of course, continues to stress couples’ economic motivations to reduce fertility. In their recent analysis of children ever born data in the 1900, 1910 and 1940-1990 IPUMS samples, Jones and Tertilt (2008) found a consistent negative relationship between fertility and “occupational income” from the earliest observable birth cohort in 1826, consistent with economic theories of the family. Other researchers have highlighted large and increasing fertility differentials between women married to men in farm and non-farm occupations, especially between women married to farmers and women married to men in professional, sales, and managerial occupations (Haines, 1992; Dribe, Hacker and Scalone, 2014).

7Urban / rural residential context has proven to be significant in many studies. Notestein drew explicit attention to the importance of urbanisation in his formulation of demographic transition theory, and his reasoning remains compelling, if somewhat imprecise about the causal mechanisms linking urbanisation with fertility decline: “The new ideal of the small family arose typically in the urban industrial society. It is impossible to be precise about the various causal factors, but apparently many were important. Urban life stripped the family of many functions in production, consumption, recreation, and education. In factory employment the individual stood on his own accomplishments. The new mobility of young people and the anonymity of city life reduced the pressure toward traditional behaviour exerted by the family and community. In a period of rapidly developing technology, new skills were needed, and new opportunities for individual advancement arose. Education and a rational point of view became increasingly important. As a consequence, the cost of child-rearing grew and the possibilities for economic contributions by children declined. Falling death-rates at once increased the size of the family to be supported and lowered the inducements to have many births. Women, moreover, found new independence from household obligations and new economic roles less compatible with childbearing (Notestein, 1953, 142).”

8Qualitative studies have stressed the importance of familial, cultural and religious change in the fertility transition. Rapid social change in the nineteenth century led to greater acceptance of the idea of smaller families, especially among native-born couples, who demonstrated greater willingness to adopt birth control methods than foreign-born couples (Smith, 1987; Degler, 1980; Vinovskis, 1976; King & Ruggles, 1990; MacNamara, 2018). Religious “free-thinkers” promoted contraceptive methods and advice, while their opponents were more likely to be drawn from the clergy (Brodie, 1994; Hacker, 1999). While the federal census did not collect individual-level data on religious affiliation, aggregate data on the size of denominations can be used to measure the religious composition of different areas (Schultz, 2006). Areas with more liberal / pietistic churches tended to have lower fertility, while areas with more conservative / liturgical churches tended to have higher fertility (Hacker, 1999; Haines & Hacker, 2011). This finding is not believed to have been the result of denominational differences in religious dogma regarding the use of contraception or the promotion of a pronatal ideology. No church prior to 1930 adopted an accepting position towards birth control and few ministers spoke publicly against the practice. Liberal churches, however, provided more positions of authority for women and promoted a form of secular individualism, while evangelical and conservative churches reinforced traditional family values and segregated gender roles, which had implications for couples’ willingness to practice family limitation strategies (Parkerson & Parkerson, 1988; Lehrer, 2004; McQuillan, 2004; Lynch, 2006; Goldscheider, 2006; Hacker, 2016). Many more examples of research attempting to quantify the relationship between fertility and cultural factors, including religion, can be found in the European literature (e.g., Derosas & Van Poppel, 2006; González-Bailón & Murphy, 2013; Daudin et al., 2016).

9Relatively little attention has been paid to the role of nearby kin in the American fertility transition. However, recent work using the Utah Historical Database found higher fertility among women with living mothers and mothers-in-law during the fertility transition (Jennings et al., 2012). This finding coheres with anthropological research emphasising the importance of kin assistance – economic and physical – to couples’ reproductive success. Couples without significant help are more likely to reduce family size, while those surrounded by kin networks will be inclined to have more children. Postmenopausal grandmothers have been identified as particularly important in this model of “co-operative breeding”. Grandmothers who are not physically frail themselves can provide knowledge and labour. In a context where most women were not working for pay outside the home, grandmothers could function as a reserve army of reproductive labour. Couples of childbearing age who reside far from their own parents must undertake childcare themselves, as formal markets for childcare services were limited until the twentieth century (Hrdy, 2009; Hawkes, O’Connell and Blurton Jones, 1989; Turke, 1988; Sear & Coall, 2011).

10Kin proximity may also influence fertility through “kin priming” (Mathews & Sear, 2013; Newson et al., 2005). With closer proximity comes more frequent social interactions with kin, and these interactions – perhaps only subconsciously – may induce people to have more children. To put it casually, kin priming is what happens when your parents ask you when you are going to have another baby. Kin priming does not stand alone from the broader conversations about reproduction that shape fertility decisions. Fecund couples are likely to be influenced by friends and neighbours, but it is likely that parents have a significant influence (Hilevych, 2018; Watkins, 1990). While kin priming and kin assistance are theoretically distinct, they tend to operate in the same direction: increasing proximity is associated with higher fertility. Sears and Coall’s recent survey (2011) indicates that the patrilineal line has a more consistent pronatal impact on fertility than maternal kin. A woman’s own family may act at times to protect women from maternal depletion – the negative impact on a woman’s own health of having additional children, or shortening birth intervals. Taken together, these conversations can still be conceptually grouped as kin priming, but the proximity of a woman’s own parents may moderate fertility.

11If the correlation between kin availability and fertility in the United States was positive and robust, the decline in kin availability over time may have played an important and unappreciated role in the fertility transition. A recent doctoral dissertation by Matt Nelson examining long-term trends in kin propinquity in the United States has documented a dramatic decline in the availability of paternal kin. In 1790, nearly 30 percent of the household heads lived within three households of another household head sharing the same surname, strongly suggesting patrilineal kinship. By 1940, the figure had declined to just 7 percent (Nelson, 2018, 103). In figure 1, we reproduce Nelson’s estimates and plot the estimates with national fertility estimates (Coale & Zelnik, 1963). The correlation may be circumstantial, of course. But the strong correlation between the two series, which is consistent with theoretical and empirical scholarship on the role of kin in fertility decisions, highlights the need for further investigation.

Fig. 1

Total fertility and patrilineal kin propinquity in the United States, 1800-1940

Total fertility and patrilineal kin propinquity in the United States, 1800-1940

12With the exception of Jones and Tertlit’s long-term study of the impact of occupational income on children ever born (2008), there are no published studies covering all years of the historical US fertility transition that rely on consistently-measured economic variables. There are no studies covering all years that include economic data with consistent measures of cultural attitudes, education, detailed nativity, city size, and the availability of kin [2]. Based on prior research, we assume that all of these factors were important and that much insight can be gained by including them together in a comprehensive model of the fertility transition. Although measurement issues make it impossible to assess the precise contributions of different factors to the fertility transition (most variables do not have a pure “economic”, “cultural”, or “familial” meaning and the cross-sectional nature of the available data do not lend themselves to constructing causal models), more comprehensive fertility models provide an opportunity to evaluate the impact of different variables at different moments during the transition, assess their contributions to the fertility transition, and examine how variables interact over time. We attempt such an integration below, focusing first on measurement issues related to kin availability, nativity, and parental religiosity.

Measurement of kin proximity

13Kin proximity is a straightforward concept, but a devil to measure historically. Inside the household, of course, we can identify couples’ co-resident kin with relationship data collected by the census. The couples we study are all women between the ages of 20 and 49, married and co-residing with their husband. We identify women’s co-resident mothers, mothers-in-law, and other females age 11 and older who might have contributed childrearing assistance. Outside the household, however, measurement is much more difficult. Direct measurement of kin proximity by scholars of the family did not begin until the 1930s (Burgess & Cottrell, 1939). Therefore, we relied on indirect measures of nearby potential kin derived from characteristics of childbearing couples and their neighbours, most importantly shared surnames. Our necessary reliance on surnames limited our ability to identify potential kin to patrilineal kin, or women who were potentially mothers-in-law of the child-bearing women we study [3]. Women identified as potential mothersin-law because of surname and birthplace similarity were required to be “ever-married”. It was nearly universal practice in the United States for widows to retain their married name, so our measure does not rely on the survival of fathers in law. In addition to surnames, we inferred the presence of potential patrilineal kin using parental birthplaces and ages, and by taking advantage of the way in which the census was taken to assess the proximity of households. In the United States the census collected information from households in essentially sequential order (Grigoryeva & Ruef, 2015; Logan & Parman, 2017; Hacker & Roberts, 2017). This household sequence is maintained in the data through a variable serial that identifies unique households within a census year. Serial numbers respect the sequence of the original enumeration that was constructed by (1) enumerating households in geographic sequence, and (2) numbering enumeration sheets in a manner that maintained this geographic order. As a result, most households in the dataset with sequential serial numbers were adjacent and those within a few serial numbers can be safely inferred as being nearby.

14Not all sequential serial numbers are adjacent, however, and not all houses that would be regarded as adjacent or proximate in space have sequential serial numbers. Thus, we are likely to be understating the frequency of proximate kin, particularly in urban areas. To understand these issues with our measure, we outline the way the US census was taken between 1850 and 1940. The formative administrative unit of the census was the enumeration district. Within enumeration districts, households were canvassed sequentially. Enumeration districts were the responsibility of a single enumerator who was responsible for carrying out the enumeration. Literally this required the enumerator to walk around the district taking the census. In the 1880 census, for example, there were 11,349 enumeration districts for a population of just over 50 million. The scale of enumeration districts varied substantially, ranging in population size from 10 to 30,000 in 1880. The largest enumeration districts were found in cities such as Chicago, New York, St. Louis and Cleveland. In cities, enumeration districts were geographically small and contiguous. The local administration of the census allowed local officials to construct enumeration districts that conformed to areas recognised by the people they were enumerating. Where the borders are known they run down major roads and along barriers to travel such as rivers or railroads. In rural areas enumeration districts also conformed to recognised neighbourhoods (Logan & Parman, 2017). For the vast majority of households within an enumeration district, households with sequential serial numbers in the data are, in fact, adjacent in physical space, and if not adjacent very close.

15In rural areas, it may be helpful for readers to have the mental image of an enumerator walking down a road, and visiting houses on either side as he (as political officials and their appointees, enumerators were nearly all men) walked down each road segment. In urban areas, it appears that enumerators were more likely to trace out block faces. They would cross the road, and enume-rate the other side of the street in a sub-sequent pass. Unfortunately, urban addresses were not collected in early censuses, and address data collected for later censuses are incomplete and messy. As a result, we were unable to use address information to infer households that may have been outside our search window of nearby households, but may still have been physically close. This will tend to bias our estimates downward, because we are not capturing all nearby kin.

16We explored several approaches to estimating the availability of nearby kin in the complete-count datasets. In preliminary analyses, we followed the existing literature and analysed only the two adjacent households (within the same enumeration district), focusing on identifying the presence of a potential motherin-law for all currently-married women of childbearing age (Grigoryeva & Ruef, 2015; Logan & Parman, 2017). We focused on mothers-in-law for four reasons: (1) they play a dominant role, along with mothers, in the theoretical literature on the “grandmother hypothesis” (Sear & Coall, 2011); (2) their greater degree of proximity to their grandchildren suggests they would have been more likely to contribute care and more intensive assistance than other patrilineal kin; (3) most mothers-in-law would have been old enough to have finished their childbearing or be near the end of their childbearing, suggesting they would have had more time to contribute assistance than other patrilineal kin; and (4) they can be identified with a high reliability. Other kin and neighbours likely contributed time and physical assistance as well, but cannot be reliably identified in the data [4]. We identified a nearby mother-in-law only if an adjacent household included a currently married, widowed or separated woman who shared the childbearing woman’s surname; had the same birthplace as the husband’s mother’s birthplace; was more than 15 years older than the husband; and was the same race as the husband. Our prior work on the 1880 census (Hacker & Roberts, 2017) extended prior scholarship by examining a wider search window around the focal household to identify additional potential mothers-in-law, although the likelihood of finding one declined rapidly with each household. Ultimately, our analysis suggested that limiting our search to ten “nearby” households, defined as the five households on either side of each focal woman, struck an appropriate balance between finding a few more potential mothers-in-law without significantly increasing the chances of misidentification or over-taxing our computing resources.

17Although we labelled our variable “potential mothers-in-law,” it is likely that a small percentage of the identified “mothers-in-law” were aunts-in-law, significantly older unmarried sisters-in-law, or other patrilineal kin. It was therefore possible for focal women to have more than one potential mother-in-law. Although a higher number of potential mothers-in-law had meaning, we decided to treat the measure as a dichotomous indicator, and thus set to one for women with one or more potential mothers-in law. The variable was set to zero for focal women with a co-resident mother-in-law, as a mother-in-law definitely identified inside the woman’s own household precludes the possibility the mother-in-law is also living in another household. If a woman had no co-resident mother-in-law and the search of ten nearby houses found no women matching the surname, age, birthplace and race criteria derived from her husband’s information the variable for potential mothers-in-law was also set to zero. We excluded from our construction of neighbouring houses any group quarters, such as prisons or hospitals or poor farms and limited our search to the nearest ten regular households. Institutional group quarters of this nature were deleted from the data before we constructed our measure of potential kin proximity. On the ground, therefore, an institution might be in between the focal household and some of its neighbours. In the data, we treated institutions as if they are not there.

Measurement of nativity

18Fertility differentials by nativity were first highlighted by nineteenth-century observers. In 1877, for example, Nathan Allen estimated that the birth rate of the foreign-born population in New England was twice that of the native-born population. He believed the result reflected a greater desire for a higher standard of living among the native-born population and, perhaps, physiological degeneration among native-born men and women related to changes in work and education (Allen, 1877).

19Modern studies have confirmed that the native-born population of New England was on the vanguard of the fertility transition (Main, 2006; Hacker, 2016). The foreign-born population lagged well behind, suggesting the persistence of customs and values opposed to the practice of birth control (Atack & Bateman, 1987; Hareven & Vinovskis, 1975). Other factors may have played a role, however, including native and foreign-born differentials in socioeconomic status, insecurities associated with minority group status, and immigrant selection factors (Goldscheider & Uhlenberg, 1969; Kahn, 1988, 1994; Forste & Tienda, 1996). Continued marital fertility decline among native-born couples, persisting high fertility rates among “old” immigrant groups, and the arrival of new immigrants from Southern and Eastern Europe with strong family systems and high fertility regimes widened fertility differentials in the early twentieth century (King & Ruggles, 1990; Gjerde & McCants, 1995; Reher, 1998; MacNamara, 2018). Morgan, Watkins and Ewbank (1994) found substantial marital fertility differentials by nativity in a small but representative sample of households of 1910 even after the controlling for age, occupation, residence, duration in the United States, and ability to speak English. Second generation couples (native-born of foreign-born parents) typically achieved fertility levels between that of native-born and first generation immigrants, suggesting a process of acculturation to American childbearing norms spanning several generations.

20The complete-count datasets, which include each individual’s place of birth (state of birth if the individual was native-born, country of birth if he or she was foreign-born) in all census years and parental birthplaces beginning in 1880, allow us to investigate many more nativity groups than prior scholars and examine changing nativity differentials between 1850 and 1930. In addition to native-born whites and blacks of native parentage, we examined the fertility of 18 first-generation nativity groups (foreign-born couples) and 18 second-generation groups (native-born couples of foreign-born parents). Because the nativity of wives and husbands were highly correlated, we decided to treat nativity as a couple-level measure. If only one partner was native-born, the nativity of the foreign-born partner was used. If both partners were foreign-born but with different nativities, we relied on the wife’s nativity. The same principle was used to measure second-generation nativity. In general, we expected that foreign-born couples originating from countries that had yet to experience the onset of the fertility transition or from countries with a later onset of the fertility transition than the United States would be less willing than native-born couples to limit their fertility. Heterogeneity within originating countries and unobserved selection criteria, however, may have resulted in the selection of low or high fertility immigrants, especially after 1880 when most European countries had commenced their fertility decline. Unfortunately, we are aware of no research on the selection of immigrants to the United States by fertility historically.

21There are a few limitations to the nativity data. Unfortunately, nativity alone is not a reliable measure of ethnicity, which was likely a more important determinant of demographic behaviour than nativity. Although it is likely that the vast majority of individuals born in Sweden were Swedish, other countries were characterised by substantial ethnic diversity. Analysis of the 1910 IPUMS sample – which includes data on language spoken, “mother tongue”, and parents’ mother tongues in addition to individuals’ country of birth – suggests that many individuals born in Austria, Hungary, Germany, Austria-Poland, Romania and especially Russia were Jewish. A substantial proportion of individuals born in Canada were Frenchspeaking Canadians, presumably from Quebec, who have been identified in other studies as having much higher fertility rates than English-speaking Canadians (Gossage & Gauvreau, 2007). Although some individuals in every census gave “Poland” as their birthplace, many Poles prior to 1918 likely reported Germany, Austria, Hungary or another country as their place of birth. In addition, boundaries and recognition of several foreign countries changed over the period covered by this study (e.g., Poland, Austria and Hungary), as did the detail with which the census instructed enumerators to report birthplaces. Although the IPUMS project attempted to create a consistent general birthplace coding system across the entire period, differences in respondent and enumerator reporting likely biased long-term comparisons of some nativity groups.

22A preliminary analysis revealed that nativity was highly correlated with the availability of kin. First-generation couples likely had fewer parents in the United States. Cultural norms about living with parents may have varied among groups, including among second-generation immigrants. In 1880, our data indicated that native-born couples of native parentage (NBNP) were 3.7 times more likely to have a potential mother-in-law residing within ten households than foreign-born couples, 65 percent more likely to have a co-resident mother-in-law, and 19 percent more likely to have a co-resident mother. Differences between NBNP couples and second-generation couples were more modest, but still significant. NBNP couples had about 37 percent more potential mothers-in-law nearby, 7 percent more co-resident mothers-in-law, and 10 percent fewer co-resident mothers than second-generation couples. To control for these differences, we interacted all 36 nativity groups with the presence of kin in our models. For space considerations, we do not show these interactions in our tables and focus our discussion on the main effects, although we discuss the general pattern and trend suggested by the interactions below.

Other independent variables

23In addition to kin and nativity, we included economic, demographic, residential, educational, and cultural variables available in each census year. During the 80 years between the 1850 and 1930 censuses the economy was dramatically transformed from agrarian to industrial with a rapidly growing service sector. The period also witnessed rapid urbanisation, increasing school attendance and the passage of mandatory school attendance laws, and religious secularisation. Among the independent variables we included in the fertility models to capture these changes, we expected spouses’ occupations to play a particularly important role in fertility differentials and in the fertility transition. Other researchers have highlighted large and increasing fertility differentials between women married to men in farm and non-farm occupations over the course of the fertility transition, especially between women married to farmers and women married to men in professional, sales, and managerial occupations (Haines, 1992; Dribe, Hacker and Scalone, 2014), reflecting changing returns to education in the industrialising economy, the costs of education and raising children, access to information, and parents’ long-term aspirations for children. We relied on the US Census Bureau’s 1950 occupational classification system – which has been applied to all US censuses by the IPUMS project – to examine the impact of the changing occupational structure on fertility.

24For school attendance we created a county-level contextual variable measuring the proportion of children age 8-14 in the county in school. Local school attendance, we assumed, is a useful proxy of parents’ perception for either the growing need to provide for their children’s education to meet the needs of a modernising economy or else a reflection of mandatory school attendance laws, although the latter tended to lag the former (Landes & Solmon, 1972; Goldin, 1999). Regardless, children in school were not in the labour force or assisting in family business or farm and represented a net cost to their parents. All else being equal, we expected fertility to be lower in areas with high school attendance and compulsory attendance laws than in areas with low school attendance, and the long-term increase in school attendance between 1850 and 1930 to be associated with fertility decline. We coded literacy as a dichotomous variable. Parental literacy might have reflected an orientation toward higher education and “white-collar” occupations for their children and greater access to printed information, including information on birth control. Because the literacy of women and their spouses were highly inter-correlated, we created a couple-level variable. We classified a couple as literate in 1850 if both the wife and husband were enumerated as being able to read and write in the 1850 census. In subsequent censuses, which recorded individuals’ ability to read and ability to write separately, we required both the wife and husband to be able to do both in order to be considered a literate couple. If one or both members of the couple were unable to read or unable to write, the couple was classified as illiterate.

25We included four dummy variables for couples’ residence location (rural, urban places with a population of less than 10,000, urban population 10,000-99,999, and urban greater than 100,000). Early demographers (e.g., Notestein, 1953) often cited urban life as incompatible with large families. Housing tended to be more expensive in cities than in rural areas, while urban children were less likely to contribute to the household economy than rural children, making urban children more expensive for their parents. At the same time, urban areas were more conducive to the spread of new ideas and attitudes than rural areas. Parents living in cities had greater exposure to information and easier access to contraceptives than parents did in rural areas. Urban areas were also associated with higher infant and child mortality rates, which affected the number of surviving children under the age of 5 in the household (Preston & Haines, 1991). We also included women’s age groups and age differentials from their spouses as demographic control variables (the latter to control for the probability of reduced sexual function, coital frequency, and fertility among much older males, which were not uncommon among women whose husbands had been previously married).

26Finally, we relied on parents’ child-naming practices to make a rough estimate of parental religiosity. As argued previously (see Hacker, 1999, 2016), parents in the United States were free to name their children as they wished. Over the course of the long nineteenth century (1780-1920), parents exhibited a remarkable trend toward more secular names. In the late eighteenth-century, for example, parents gave biblical names to more than 65 percent of their male children. By 1920, the percentage had fallen to less than 18 percent (Hacker, 1999). Contemporary observers lamented these changes and contended the increasing choice of secular names reflected declining parental religiosity. With some qualifications, we agree. Rather than assuming parents choosing a higher proportion of biblical names for their children held more deeply felt religious beliefs than parents choosing a higher proportion of secular names, we believed it more likely that parents relying on biblical names were less open to sources outside of religion for authoritative positions on various topics, including contraception and abortion, consistent with “neo-secularisation” theory (Chaves, 1994; Yamane, 1997; Moore, 1989). Some measurement error is inevitable, of course. To the extent that the measure imperfectly captures parental religiosity, coefficients will be biased downward. To be included in the subsequent analysis, all couples were required to have a least one child with a valid given name [5]. In all regression models, couples’ nativity was interacted with the proportion of children given biblical names to control for the possibility of couples emigrating from countries with significant naming restrictions and different naming traditions (Hacker, 1999; 2016). Given space considerations, however, we focus our discussion of the main effect in the NBNP population.

Methods and descriptive results

27Our analysis is based on Poisson regressions of the number of own children less than age 5 as the dependent variable. Because we lacked information on the number of children who may have died in the five years prior to the census, the variable is more precisely a measure of net marital fertility or marital reproduction. Differentials in net fertility may reflect, in small part, differentials in infant and child mortality rather than differentials in marital fertility. However, a recent analysis of occupation differentials in fertility in the 1910 census – which recorded the number of children ever born and number of children surviving – obtained nearly identical results when the analysis was based on the number of living children under age 5 and the number of children born in the previous five years (Hacker, 2019, table 5), suggesting that the number of living children is a good proxy of couples’ marital fertility, at least in 1910. We constructed models for each census year (1850, 1880, 1910, and 1930). For all models, the universe was defined as currently-married women age 20-49 with spouses present in the household and with one or more children with valid names. In 1850, the universe was further restricted to the free population. Unfortunately, the nation’s 3.2 million enslaved inhabitants – approximately 88 percent of the black population – were enumerated on a special slave schedule with limited information and could not be included. After further restricting the universe to women with non-missing information, our study population included more than 31 million women. All models included fixed effects at the State Economic Area level (SEA), an aggregation of two or more contiguous counties identified by the 1950 census as sharing similar economic characteristics, to control for unobserved heterogeneity. With the territorial growth of the United States between 1850 and 1930, the number of SEAs increased. In 1850, there were 366 SEAs containing an average of about six thousand childbearing women; in 1930 there were 455 SEAs with an average of more than 30 thousand women in each.

28Means for variables in the regression models are shown in table 1 for each census year. The means for the dependent variable fell from 1.178 children under age 5 per married woman in 1850 to 0.709 in 1930. Because infant and child mortality also declined, the actual decline in fertility was more pronounced. According to Coale and Zelnick’s estimates of the total fertility rate for the white population shown in figure 1 (1963), fertility declined 54.8 percent between 1850 and 1930. In 1850, 7.8 percent of women had a potential mother-in-law living within the closest 10 households, while only 4.1 percent had one in 1930. Unsurprisingly, the number of co-resident females age 11 and older also fell as fertility declined, from 0.86 in 1850 to 0.72 in 1930. The percentage of currently-married women with co-resident mothers and mothers-in-law was low – ranging from about 3 to 4 percent – with little apparent trend. There were proportionately more potential mothersin-law in nearby households and co-residing mothers and mothers-in-law in eastern census regions, consistent with known patterns of migration of younger generations to the western frontier and the preference of older generations to remain in or near their long-term homes (not shown in table). School attendance among children age 5-17 between 1850 and 1930 increased from 58.1 to 79.6 percent while the proportion of children biblically named fell from 46.8 to 20.6 percent. Literacy rates were already high in 1850 – 83.0 percent of couples included both a literate husband and a literate wife – but increased an additional 9.6 percent to 92.6 percent by 1930.

29As expected, the descriptive results also highlight a shift in the occupational structure, with falling percentages of spouses engaged in agricultural occupations between 1850 and 1930 and rising percentages in professional, industrial, and service occupations. In 1850, half of women’s husbands were either farmers, farm operatives, or farm labourers. By 1930, the percentage had fallen to 23.1. Concomitant with the shift in occupational structure was a shift toward urban residences and larger cities. In 1850, 84.6 percent of couples in the study lived in rural areas and 5.3 percent lived in a large urban city with a population of more than 100,000. Eighty years later, in 1930, the rural percentage was 44.7 percent while the percentage of couples living in a large city had increased more than five-fold to 29.6 percent.

30Finally, changes in immigration to the US are reflected in the changing proportions of couples with different nativities [6]. The three largest nativity groups in 1850 – individuals born in Germany, Ireland, and Great Britain – declined in proportions between 1850 and 1930 while new immigrant groups – noticeably individuals born in Russia, Italy and Poland – increased. The combined percentage of couples who were foreign-born increased from 18.4 percent in 1850 to 27.7 percent in 1910 and then declined to 23.5 percent in 1930 following the passage of immigration restriction legislation in the 1920s. With increased immigration the percentage of couples who were second generation increased, from 13.6 percent in 1880 to 25.5 percent in 1930.

Poisson regression analysis

31The results of the Poisson regressions are shown in table 2. Because we are reporting model results for the entire population and rely on millions of observations, we do not present sample statistics and focus solely on the size and sign of the coefficients (although not germane to population estimates, nearly all coefficients are “statistically significant” at the 0.001 level). Results are generally consistent with expectations. Women married to men in non-farm occupations had fewer children less than age 5 in the household than the reference group of women married to farmers. Exponentiating the coefficient for women married to men with professional and technical occupations in 1930, for example, indicates that these women had 18.3 percent fewer children than women married to farmers. The largest differentials were between women married to farmers and women married to men in service occupations or in one of the three “white collar” occupation groups: professionals and technical workers; managers, officials, and proprietors; and clerks and sales workers.

Tabl. 1

Means of variables in the models

Means of variables in the models

Notes. Parental birthplaces not available in 1850 census. Black population in 1850 excludes slaves. Interactions between nativity variables and potential childrearing assistance variables and nativity and biblical naming of children not shown. Universe includes all currently married women age 20-49 with spouses present. 1850 population excludes slaves.

32Residence location was an important factor in couples’ net childbearing. All else being equal, couples living in urban areas had fewer children under the age of 5 than couples living in rural areas in all census years; the difference grew greater over time. Compared to couples living in rural areas, couples in urban areas had 7-8 percent fewer children in 1850 and 10-13 percent fewer children in 1930. There was little substantive difference between couples living in large urban and small urban places in 1850 – all else being equal, couples living in large urban places had 0.7 percent more children in 1850 than couples living in small urban areas. By 1930, however, couples in large urban areas had about 3 percent fewer children than couples in small urban areas. Unsurprisingly, older women had fewer children than the reference group of married women age 20-24. The exponentiated coefficients also reflect the well-known increase in parity-dependent control over the course of the fertility transition. In 1850, women age 40-44 bore 43.9 percent fewer children than the reference group of women age 20-24; by 1930 women in this age group had 75.6 percent fewer children.

33The results indicate that couples living in counties with higher percentages of children in school had fewer children than couples with lower percentages in all census years. The substantive impact was relatively modest in 1850 and 1880, when couples living in a county with 25 percent more students in school had less than one percent fewer children under age 5, and substantively greater in 1910 and 1930, when the same difference was associated with about 4 percent fewer children. Literate couples had 5-7 percent fewer children in 1850, 1880, and 1910, and 10.8 percent fewer in 1930. Only 7 percent of couples in 1930, however, had an illiterate husband or wife. Parental religiosity, as proxied by parents’ choice of biblical names for their children, appears to have been a significant obstacle to practice of marital fertility control. NBNP couples who relied solely on biblical names for their children had 0-7 percent more children than couples relying solely on secular names, depending on the census year, with the greatest differences later in the fertility decline. The true impact of parental religiosity was likely larger. As previously noted, the child naming variable is believed to be an imperfect proxy of parental religiosity, and therefore understates its importance.

34As expected, NBNP women with potential mothers-in-law nearby had more children, all else being equal, than women without potential mothers-in-law nearby, consistent with these couples receiving greater assistance with childrearing and other types of assistance, and also greater exposure to “kin priming”, both of which have been associated with higher fertility. The main effect was substantively modest in all census years, however. In 1910, for example, the exponentiated coefficient indicates that the presence of a potential mother-in-law nearby was associated with 0.5 percent higher fertility for NBNP couples, all else being equal. The results for the interactions between the 36 nativity groups (not shown for space reasons) and nearby potential mothers-in-law varied, but indicated that the presence of potential nearby mothers-in-law had a modestly lower impact for most foreign-born groups relative to the NBNP in 1850 and 1880 and a modestly higher impact relative to the NBNP in 1910 and 1930. The substantive impact of nearby mothers-in-law is likely biased downwards, however, by our failure to identify all potential mothers-in-law living nearby. Many married women no doubt received assistance from potential mothers-in-law living outside our search window of the ten closest households. This was especially true for women living in urban areas, where residences separated by much more than five households may have in fact been quite close to each other. A separate model for the native-white population of native parentage living in rural areas (results not shown), yields somewhat higher, although still modest, coefficients (in 1910 rural areas, nearby mothers-in-law were associated with 1 percent higher fertility). The coefficient for nearby mothers-in-law is also biased downwards by our failure to identify all potential nearby childrearing assistance outside the household, most notably childbearing women’s own mothers and other nearby relatives (sisters, aunts, and some male relations).

35Co-residence with mothers-in-laws and mothers, however, was associated with lower fertility, as was the co-residence of other older females, suggesting that the economic and childrearing assistance these women presumably provided to childbearing couples was counterbalanced by other factors not measured in the model. Unobserved selection factors may also bias the results. If the propensity of mothers and mothers-in-law to live with their offspring was correlated with disability, poor health or poverty, for example, co-resident parents may have represented an additional burden for couples in the model, not a source of assistance. It is also possible that families with more children had less space to accommodate co-residing kin, reversing the potential cause and effect. An older parent deciding among one of two possible children with which to core-side with might have decided to move in with the child having fewer children, and thus more room. The same selection factors may also bias results for nearby mothers-in-law. Longitudinally-linked census samples – now in construction at the Minnesota Population Center – will allow us to untangle potential selection biases by observing the impact of changes in living arrangements with changes in fertility. Future work will also examine, where possible, the potential role of nearby siblings, other kin, and the age of co-residing mothers, mothers-in-law and nearby kin, which may prove to be an important proxy of their ability to provide physical assistance [7].

36The model’s results were consistent with the hypothesis that nativity played a major role in fertility differentials. Fertility differences between first generation couples and native-born couples of native parentage (NBNP) were very large, especially near the onset of the fertility transition between 1850 and 1880, consistent with an earlier onset of fertility decline in the native population. In 1880, couples born in Germany, Russia, Sweden, Ireland, Austria, Norway, Denmark, Switzerland, Poland and the Netherlands had fertility rates at least 30 percent higher than that of NBNP couples, all else being equal [8]. Foreign-born women achieved higher fertility rates than the NBNP population, it should be noted, despite the fact that fewer nearby kin were available for assistance. Fertility differentials between the foreign-born population and NBNP population narrowed significantly in the later years of the fertility decline with some groups (Germans, Russians, Swedes, Austrians, Hungarians and French) achieving lower fertility than the NBNP by 1930. Among the 18 different nativity groups examined, only two groups continued to experience fertility rates in 1930 that were in excess of 30 percent greater than the NBNP rate – couples born in Ireland and couples born in Mexico – while the rate for couples born in Italy was 29 percent higher. Interestingly, all three groups had significant Catholic majorities, a marked contrast to the largely Protestant NBNP population and most other immigrant groups. The Protestant-Catholic differential in fertility was identified as significant by early twentieth-century demographers, widened during the baby boom, and was identified as a critical determinant of contraceptive use and behaviour from the 1950s through the “end of Catholic fertility” in the 1970s (Westoff & Jones, 1979). For most groups, fertility differentials between NBNP couples and second-generation couples were less than one-half of that of the first generation in 1880 (see table 2). The fertility differential between the NBNP and most second-generation groups had largely disappeared by 1930. As a result, the foreign-born population, which had helped slow the pace of the fertility transition in late nineteenth and early twentieth centuries, now were major participants in the decline. By 1940, or a few years earlier, the US population experienced nearly sub-replacement fertility rates for the first time, with the white population reaching a total fertility rate of 2.23 in 1940 and the black population a total fertility rate of 2.87.

Tabl. 2

Poisson regression of recent net marital fertility (own children less than age 5 in the household)

Poisson regression of recent net marital fertility (own children less than age 5 in the household)

Notes. Parental birthplaces are not available in 1850 census. The black population in 1850 excludes slaves. Interactions between the 36 nativity groups and the nearby availability of potential mothers-in-law and between the 36 nativity groups and biblical naming of children not shown. Universe includes all currently married women age 20-49 with spouse present.

Contribution of independent variables to the fertility decline

37In table 3 we make a rough assessment of the contribution of the independent variables in the models to the decline of net marital fertility (as measured by our dependent variable) in the three periods spanning the inter-census years 1850-1880, 1880-1910, and 1910-1930, and an estimate over the entire 1850-1930 span. The estimates were derived using the average of the coefficients from the beginning and end years, the means of the beginning and ending values for all variables (with the exception of the variable being assessed), and by allowing the value of the assessed variable(s) to vary from the mean value at the beginning of the period to the mean value at the end of the period [9]. The percentages shown are the percentage of the observed decline in the number of children under age five explained by the variable(s). In order to be substantively significant to fertility decline, the variable (or group of variables) had to be substantively correlated in cross-sectional census regressions and to have experienced a significant change in level over the period consistent with the trend in fertility. Few variables met this standard, suggesting possible omitted variables and other types of bias. We have, for example, no measures of the supply of birth control technology and diffusion of information on various methods, which likely played significant roles in the fertility transition.

Tabl. 3

Predicted effect of change in selected independent variables on number of own children less than age 5 in the household and as percentage of the observed decline, by period

Predicted effect of change in selected independent variables on number of own children less than age 5 in the household and as percentage of the observed decline, by period

Notes. Values for occupational structure, nativity, residence, and age structure are the combined effect of all variables in those categories. A negative percentage indicates that the independent variables worked against fertility decline in the period. If couples’ nativities had remained constant between 1880 and 1910, for example, the model predicts that fertility would have declined another 4.9 %. n.a. = not applicable.

38A few variables do stand out as significant, however, in both their consistent cross-sectional impacts and in their likely contributions to the fertility decline. Unsurprisingly, the shift in occupation structure explained some portion of the decline in fertility in each period: 5.9 percent of the decline in fertility between 1850 and 1880, 3.8 percent of the decline between 1880 and 1910, 2.4 percent of the decline from 1910 to 1930, and 3.5 percent of the decline overall. Increased school attendance accounted for 2.7 percent of the decline between 1850 and 1930, while increased literacy accounted for 1.2 percent of the decline. The increased tendency of married couples to live in urban areas, especially larger urban areas, explained 6.1 percent of the observed decline in net fertility between 1850 and 1880, 4.0 percent of the decline between 1880 and 1910, 2.6 percent of the decline from 1910 to 1930, and 4.1 percent of the decline overall. The greater contribution of urban residence to the decline in the earlier interval was likely the result of increasing child mortality in urban areas prior to the public health movement of the late nineteenth century and the onset of the mortality transition (Preston & Haines, 1991). Secularisation – as proxied by parents’ declining use of biblical names – contributed modestly to the decline in most periods, explaining just 1.4 percent of the overall decline. Changes in nativity worked against the fertility transition in each period, with the arrival of high fertility immigrant couples in the late nineteenth and early twentieth centuries, although fertility decline among the foreign-born population accelerated after 1910. Among the most substantively important factors for women in the model was the shift in the age structure of currently-married women, likely caused by increases in the mean age at marriage and the greater survival of women through their childbearing years (Haines, 1996). Compared to the age structure of married women in the nineteenth century, the age structure of married women in the twentieth century was older and therefore less conducive to high marital fertility. We should bear in mind, however, that although the shift in age structure of child-bearing women contributed to the decline in the number of children under age 5 for women in the model, changes in the population’s age structure does not play a role in the long-term decline of total fertility rates in the United States from 7.0 in 1835 to 2.1 in 1935 cited earlier and shown in figure 1, which are based on age-specific fertility rates for synthetic cohorts of women and therefore insensitive to changes in age structure. A similar table for native whites of native parentage living in rural areas (not shown) indicates similar contributions by most variables to the decline.

39Despite the long-term decline in the availability of nearby mothers-in-law, our analysis suggests that the long-term decline in childrearing assistance from mothers-in-law played a very minor role in the fertility transition. In all periods, changes in the overall proportion of couples with a nearby mother-in-law explained less than 0.2 % of the fertility decline. We stress, however, that our measure of kin availability – based as it is on only the presence of potential mothers-in-law in the nearest 10 households – is very conservative. The true coefficient, if kin availability could be precisely measured, was likely higher. Also of note, our measure of potential mothers-in-law declined from only 8 percent of all couples in 1850 to 4 percent of couples in 1930, a net change of 4 percent. Nelson (2018), however, documents a decline in all nearby patrilineal kin (+/− 3 households) from approximately 18 percent to 7 percent during the same period. If the true decline in the availability of mothers-in-law was similar and the true coefficient higher, and if maternal kin and other paternal kin played a similar role and trend, declining kin availability may have rivalled the importance of other factors commonly accepted as significant in the fertility transition. Work on linked census datasets currently in progress at the Minnesota Population Center will allow us to obtain better estimates of the impact of kin on couples’ reproductive success (Hacker et al., 2019).

Conclusion

40Research on the US fertility transition typically ignores the potential contribution of cultural and familial influences (important exceptions include King & Ruggles, 1990; Guinnane, Moehling and O’Grada, 2006; Gjerde & McCants, 1995). In this paper we relied on the new IPUMS complete-count 1850, 1880, 1910, and 1930 census microdata databases to study the role of economic, social, cultural, and familial factors in the fertility transition, including the investigation of whether proximity to nearby kin influenced couples’ fertility behaviour. By examining the surname, age, and sex of members in adjacent households, we were able to construct a measure of potential mothers-in-law for all women of childbearing age in the dataset. We also constructed measures indicating the coresidence of mothers, mothers-in-law, and other females age 11 and older. The results indicated that while proximity of mothers-in-law had a very modest positive impact on women’s fertility – consistent, however, with hypotheses that the availability of assistance is positively correlated with fertility – co-residence with mothers, older daughters, and other women had a negative impact. This negative impact, however, is likely biased by unobserved selection biases, and the modest impact of kin outside the household is likely an overly conservative estimate. Future studies will link individuals across censuses – allowing selection biases to be minimised – and construct wider measures of kin availability.

41We also examined the impact of nativity and a proxy of parental religiosity on fertility. Both proved to be significant correlates of marital fertility. Parents who chose a higher proportion of biblical names for their children had higher fertility rates than parents who relied on secular names, suggesting a positive relationship between parental religiosity and marital fertility. Couples’ nativity exerted the strongest influence of all independent variables in the model. Foreign-born couples from most of the nations examined had dramatically higher net marital fertility rates than native-born couples of native parentage. These large differentials, which remained substantively significant after controlling for economic, demographic and other suspected covariates, suggest that cultural attitudes played a major role in couples’ childbearing decisions. All else being equal, native-born couples with native-born parents proved more willing to act on incentives to reduce their fertility, while foreign-born parents proved less willing. In most cases, fertility differentials between native-born couples of native parentage and second-generation couples were less than half of the differentials estimated for first-generation couples. With the exception of three largely Catholic nativity groups, however (couples born in Ireland, Mexico, and Italy), nativity differentials had narrowed or reversed by 1930.

42Unsurprisingly, the changing occupational structure and increasing urbanisation of the population over the period studied provide the most consistent and substantive contribution to fertility decline. Early researchers stressing the importance of urbanisation, industrialisation and rising incomes to the fertility transition were not wrong. Overall, however, we believe our results demonstrate the need for more inclusive models of fertility. Too often, prior research has focused exclusively on economic factors including land availability, occupational structure and women’s earning potential. Although economic motivations were clearly important, couples’ fertility decisions depended on a host of factors, including proximity to kin, nativity, and religiosity. We conclude that failure to consider the role of culture and the role of the family result in an incomplete understanding of the couples’ decisions on the number and timing of their children.