Which best describes life expectancy and marriage in the seventeenth century Chesapeake?

Hist Methods. Author manuscript; available in PMC 2011 Apr 1.

Published in final edited form as:

PMCID: PMC2885717

NIHMSID: NIHMS205286

Abstract

This article constructs new life tables for the white population of the United States in each decade between 1790 and 1900. Drawing from several recent studies, it suggests best estimates of life expectancy at age 20 for each decade. These estimates are fitted to new standards derived from the 1900–02 rural and 1900–02 overall DRA life tables using a two-parameter logit model with fixed slope. The resulting decennial life tables more accurately represent sex-and age-specific mortality rates while capturing known mortality trends.

Keywords: mortality, life table, nineteenth century, United States, demography, demographic history

Life tables summarize the effects of age-specific mortality rates on a real or synthetic cohort. In addition to their descriptive value, life tables are an indispensible tool for demographers, with many applications in the study of mortality, fertility, migration, and population growth. Life tables are often used in conjunction with indirect estimation methods for the study of populations covered by a census but lacking a reliable vital registration system, such as many populations in developing countries and populations in the past. Life tables, for example, can be used to estimate vital rates from census age distributions and are required in own-child fertility analysis (see United Nations 1983, for a description of commonly-used indirect methods).

Demographic historians of the nineteenth-century United States depend heavily on life tables and indirect estimation methods. Although the federal government conducted a census every ten years, it did not implement a vital registration system until the start of the twentieth century (the system was not complete until 1933).2 D As a result, the timing and contours of the demographic transition in the United States are less precisely known than that in nations such as England and Australia, which had comprehensive birth and death registration by the mid nineteenth century (Jones 1971; Woods 2000). Despite this limitation, demographic historians have been able to estimate annual and age-specific birth rates, net migration rates, and cohort trends in life cycle experiences as far back as the early nineteenth century using census data, life tables, and indirect methods (Yasuba 1962; Coale and Zelnik 1963; Kuznets 1965; Uhlenberg 1969; McClelland and Zeckhauser 1982; Tolnay, Graham and Guest 1982; Ferrie 1996; Hacker 2003).

Unfortunately, the results of these studies depend on a small number of life tables, which suffer from limited geographic coverage, limited temporal coverage, and a variety of source-based problems. The earliest life tables rely heavily on data from Massachusetts, a small state in the Northeast characterized by much higher levels of urbanization, industrialization, and immigration and much lower levels of nuptiality and fertility than the nation as a whole. Given the high short-term variability in mortality rates that was characteristic of the nineteenth-century United States, it is also unclear whether life tables based on a single year of data can be used to represent mortality in a year other than that for which it was constructed. The failure of existing life tables to capture suspected long-term trends in mortality is perhaps their most significant limitation. With just a handful of life tables covering the entire nineteenth century, researchers have been forced to make crude assumptions about long-term mortality trends to conduct their analyses. As discussed in more detail below, recent research indicates that earlier assumptions of long-term mortality decline are in error. Mortality increased significantly in the mid nineteenth-century United States before beginning its long-term decline.

This article constructs new life tables for the white population of the United States in each decade between 1790 and 1900. The first part of the article reviews research on the level and trend of nineteenth-century mortality. Drawing from several recent studies, it suggests best estimates of male life expectancy at age 20 for each decade. The second part of the article investigates sex differentials in mortality and suggests best estimates of female life expectancy at age 20 for each decade. The third part of the article reviews research on the age pattern of male and female mortality. The results indicate that age-specific mortality rates in the nineteenth century did not match the two most frequently cited standards: the “West” model of the Princeton regional model life tables or the 1900–02 Death Registration Area (DRA) life tables for the United States. It concludes, however, that the life tables constructed for the rural part of the 1900–02 DRA likely approximates the age pattern of nineteenth-century mortality. Finally, the fourth part of the article fits the decennial estimates of life expectancy at age 20 to new standards derived from 1900–02 rural and overall life tables using a two-parameter logit model with fixed slope. The resulting decennial life tables, it is argued, more accurately represent sex-and age-specific mortality rates while capturing known mortality trends.

The Level and Trend in Nineteenth-Century Mortality

Table 1 shows life expectancy and infant mortality estimates from selected United States life tables in the period between 1789 and 1902 by year of publication (for a more complete listing, including life table summaries for selected cities, see Haines 1998). The tables were constructed from a wide variety of sources, including local bills of mortality, state and national death registration data, census data, family genealogies, and biographical data on special populations such as legislators and college graduates. Table 1 also shows the sex mortality differential at age 20, defined as the female life expectancy at age 20 minus male life expectancy at age 20.

Table 1

Life Tables Estimates of Infant Mortality and Life Expectancy at Selected Ages, United States 1798–1901

Edward Wigglesworth (1793) constructed the first United States life table using Bills of Mortality for 35 New England towns in the late eighteenth century. Ezekiel B. Elliott (1858), John S. Billings (1885), Samuel W. Abbott (1898), and James W. Glover (1921) relied on death registration data in Massachusetts—the first state to implement a death registration system—to calculate life tables for selected years in the late nineteenth century.3 Levi Meech (1898) constructed the first national life table. The lack of national death registration data forced Meech to rely on an indirect approach. He estimated cohort declines from the 1830–1860 federal censuses, made adjustments from immigration data to account for the lack of a closed population, and used retrospective mortality data published by the 1860 census to establish the age pattern of death (1898, 255–59). The creation of a national death registration area (DRA) in 1900 greatly facilitated the creation of life tables. James Glover’s 1900–02 DRA life tables (1921) relied on registration data from the ten states and the District of Columbia that comprised the nation’s original DRA. These tables have been widely used by researchers to represent the level and age pattern of mortality in the United States at the turn of the twentieth century.

Two studies conducted in the mid twentieth century have been widely cited as representative of nineteenth-century mortality. A. J. Jaffe and W. I. Lourie (1942) relied on death registration data collected by 44 New England towns, several mid-sized cities, and a few larger cities to construct life tables for the period 1826–35. The results indicated large differentials in life expectancy between rural areas and large urban centers, with life expectancy at birth almost 15 years higher in the selected towns than in the large cities of Boston, New York, and Philadelphia. Paul H. Jacobson’s 1849–50 life tables were based on retrospective mortality data collected by the 1850 census. Jacobson confined his analysis to data collected for Massachusetts and Maryland, reasoning that an arithmetic mean of their age-specific death rates would approximate those for the nation as a whole (1957).

The life table estimates in Table 1 are sorted by year of publication to emphasize our relatively recent knowledge about nineteenth-century mortality. Researchers requiring life table data prior to the late 1970s were limited to a handful of tables, which led to great uncertainty about mortality trends. Inferring mortality trends in the early nineteenth century from existing life tables was especially problematic. Warren S. Thompson and P. K. Whelpton calculated a slow decline in the crude death rate from 27.8 per thousand in the late eighteenth century to 21.4 per thousand in 1855 by interpolating between the Wigglesworth and Elliott life tables (Thompson and Whelpton 1933, 230–31). Reviewing the more recent evidence available to them in the late 1950s, Conrad Taeuber and Irene B. Taeuber found no conclusive evidence of mortality decline in the first half of the nineteenth century (1958, 269). Yasukichi Yasuba saw evidence of mortality increase in the few decades preceding 1860 associated with increasing urbanization and declining sanitary conditions (1962, chapter 3). Richard Easterlin, in contrast, argued that increasing per capita income more than offset the negative impact of urbanization and cited life expectancy estimates from the Wigglesworth and Jacobson life tables as evidence of significant mortality decline (1977).

Most early observers agreed that the latter half of the century was characterized by substantial mortality decline, although opinions differed about the date of its onset. Taeuber and Taeuber thought the evidence suggested an “almost continuous” decline in mortality beginning about 1850 (1958, 269). To conduct their classic study of long-term trends in white birth rates, Ansley Coale and Melvin Zelnik assumed a linear decline in mortality between Jacobson’s 1849–50 life tables and the 1900–02 DRA life tables (1963). In separate analyses based on Simon Kuznets’ census-survival estimates of crude death rates (1965), however, Edward Meeker (1972) dated mortality decline after 1880, when the public health and sanitation movement became more effective, while Robert Higgs (1973) observed a decline in rural areas from the 1870s.

Beginning in the 1970s, new research considerably clarified our understanding of nineteenth-century mortality. Much of the new research was critical of earlier studies. In a series of articles, Maris Vinovskis (1971, 1972, 1978) evaluated the Wigglesworth, Jaffe and Lourie, Elliott, and Jacobson life tables, all of which relied on data from Massachusetts. Although the Wigglesworth life table suggested a reasonable estimate of life expectancy at birth, Vinovskis observed that Wigglesworth lacked adequate data on the age distribution of the towns, which required adjustments amounting to “little more than intelligent guessing” (1971, 589). Vinovskis also noted that the towns covered by the Wigglesworth life table were not representative of other New England towns in important characteristics, including their relative affluence and degree of urbanization, making it difficult to evaluate the table’s respresentativeness. Vinovskis faulted Jaffe and Lourie for relying on data from many small towns with under-registered deaths, thus overestimating the significance of the rural-urban differential in mortality and understating the overall level of morality (1972, 204–5). Elliott, Vinovskis argued, erred in the opposite direction. To avoid including places with deficient record keeping, Elliott eliminated towns with a crude death rate of less than 16 per thousand. In doing so, however, Elliott likely removed towns whose true death rate was lower than 16 per thousand and thus overstated the true level of mortality. Vinovskis also contended that Elliott’s reliance on just one year of mortality data was problematic, given the era’s high short-term variability in death rates (1972, 208–10). Finally, Vinovskis demonstrated that Jacobson failed to consider contemporary critiques of the 1850 census of mortality, which noted that deaths were unevenly registered, and failed to consider that the census was taken during a cholera epidemic, resulting in a likely overestimation of mortality despite the under-registration of deaths. Given these critiques, it is no surprise that Vinovskis strongly cautioned against inferring mortality trends from the Wigglesworth, Jaffe and Lourie, Elliott, and Jacobson life tables. Drawing from bills of mortality and state registration reports, he concluded that there was little trend in Massachusetts mortality during the first half of the century (1978).

Meech’s life table also received an extensive critique. In a detailed reconstruction and analysis, Michael R. Haines and Roger C. Avery noted that Meech was forced to make a number of assumptions to construct his life table, including the questionable assumptions that the underenumeration of deaths in the census and the required adjustment of gross to net migration were independent of age. As a result, Haines and Avery concluded that the Meech life table likely underestimated infant mortality and overestimated early childhood mortality, although it gave reasonable results overall (Haines and Avery 1980).

Finally, a number of researchers have cautioned against inferring national mortality patterns from life tables constructed for Massachusetts and the 1900–02 Death Registration Area (Easterlin 1977, 133; Condran and Crimmins 1979, 1; Preston and Haines 1991, 49–50; Haines and Preston 1997). Although these tables were based on relatively well reported death registration data4 —and are thus reasonably accurate descriptions of the level of mortality and sex- and age-specific mortality patterns in those areas—they are unlikely to be representative of the national population. Table 2 compares the population of Massachusetts, the 1900–02 DRA, and the overall United States in 1850 and 1900 using data from the 1850 and 1900 IPUMS samples (Ruggles et al. 2009). Massachusetts was much more urban than the rest of the nation, had a proportionately larger and more rapidly growing foreign-born population, and had a much lower proportion of its labor force engaged in agriculture (the state was one of the first to industrialize in the early nineteenth century). Moreover, Massachusetts enjoyed one to the best public health systems in the nation and was the leading state in the employment of women in the labor force and in the fertility transition. Massachusetts women age 20–49, for example, had an average of just 1.5 co-residing own-children in 1850 and 1.3 in 1900, suggesting fertility rates approximately one-third lower than that of the nation as a whole. Thus, although Massachusetts has the best available mortality data for the nineteenth century, its level, trend, and age pattern of mortality are unlikely to be representative of the United States as a whole.

Table 2

Comparison of Selected Characteristics of Massachuetts and the Original Death Registration Area States with the United States as a Whole, 1850 and 1900

Table 2 also indicates that the population of the 1900–02 DRA was not representative of the nation. The initial DRA included the six New England states, New York, New Jersey, Michigan, Indiana, and the District of Columbia. Although the 1900–02 DRA was much larger than the state of Massachusetts—representing about 26.2 percent of the nation’s population in 1900 compared to just 3.7 percent for Massachusetts—it varied from the rest of the nation in similar, if less dramatic, ways. The DRA was more urban than the United States as a whole and its population included a higher proportion of foreign born residents and a lower proportion of agricultural workers. Women in the DRA had an average of 20 percent fewer co-residing children in the household than women in the nation as a whole. It is noteworthy, however, that differences between the rural parts of the 1900–02 DRA and the rest of the nation were less extreme. Rural parts of the DRA included about the same proportion of foreign born residents and workers engaged in agriculture. Fertility rates in rural areas of the DRA were much closer to the national average.

Fortunately, just as confidence in the representativeness and accuracy of existing life tables was falling, new research significantly enhanced our understanding of nineteenth-century mortality trends. Beginning with Michael Haines’ analysis of the United States censuses of mortality (1979) and Kent Kunze’s (1979) and Robert W. Fogel’s (1986) demographic analyses of family genealogies, life expectancy estimates have accumulated for each decade of the nineteenth century. Clayne Pope’s study of family histories (1992) is perhaps the most significant contribution for the first half of the century while Haines’ construction of life tables (1998) for the white, black, and overall populations is the most important work for the last half of the century, although important research has also been published by R. S. Meindl and A. C. Swedlund (1977), Gretchen A. Condran and Eileen Crimmins (1979, 1980), Eileen Crimmins (1980), Daniel Scott Smith (1982, 2003), Gretchen A. Condran and Rose A. Cheney (1982), Rose A. Cheney (1984), Stephen Kunitz (1984), Gretchen A. Condran (1987), Richard Steckel (1988), Barbara J. Logue (1991), Eric Leif Davin (1993), Alice Kasakoff and John Adams (1995; 2000), Joseph Ferrie (1996, 2003), Antonio McDaniel and Carlos Grushka (1995), J. David Hacker (1997), John E. Murray (1997, 2000), Chulhee Lee (1997, 2003), Susan I. Hautaniemi, Alan C. Swedlund and Douglas L. Anderton (1999), Douglas L. Anderton and Susan Hautaniemi Leonard (2004), and Jeffrey K. Beemer, Douglas L. Anderton and Susan Hautaniemi Leonard (2005).

Several of the newer studies—including those by Haines, Ferrie, and Condran and Crimmins—have relied on retrospective mortality data collected by the Census Office/Bureau of the Census between 1850 and 1900. Beginning In 1850, census marshals were instructed to record the name of every person in the household who died in the year prior to the census, as well as their age, sex, race, marital status, occupation, and cause of death. The collected data were tabulated and published in separate mortality volumes. These tabulated data appear to be tailor-made for the construction of life tables: the number of deaths at each age and sex can be used as the numerator in the calculation of age-specific death rates while the denominator for the mid-year population in each age group can be obtained (with some adjustment for population growth in the preceding year) from the regular census enumeration. Census officials, however, immediately discerned that the mortality data were underreported by approximately 40 percent. Life tables could only be constructed by making large (and ultimately unknowable) adjustments to the number of deaths reported at each age (e.g., see E. B. Elliott’s “approximate” life table for the 1870 population (1874)). Differential mortality could be examined only by assuming no differentials in undercounts. Census officials clearly believed, however, that the undercount varied by region, urban/rural residence, and between long and recently settled states. J. D. B. DeBow, Superintendent of the 1850 census,, contended that state differentials in death rates “show not so much in favor of or against the health of either, as they do, in all probability, a more or less perfect report of the marshals. Thus it is impossible to believe Mississippi a healthier State than Rhode Island…” (1855, 8). Despite this disappointment, and the urging of some census officials to drop the expensive undertaking, the mortality information was deemed useful enough to continue its collection and publication through the 1900 census. More questions were added and, beginning in 1880, the information was supplemented with death records from states with available registration data (Condran and Crimmins 1979).

Retrospective mortality data were undercounted for several reasons. Most obviously, solitary households left no one behind to report the death to an enumerator. The death of a household member of a larger family, especially the household head, often led to the dissolution of the household. Respondent error also led to undercounting. Deaths of infants and the elderly were underreported, and deaths occurring 6–12 months prior to the census enumeration were less likely to be reported than deaths occurring 0–6 months prior to the count (Condran and Crimmins 1979; Ferrie 2003). In an early comparison of death reporting between the 1880 census of mortality and the early death registration states of Massachusetts and New Jersey, J. S. Billings observed that “the proportion of deaths omitted in the enumerators’ returns increases in a tolerably regular manner as we go back in time from the date of enumeration.” Billings calculated that census undercounting of deaths in the 1880 census increased from about 17 percent of all registered deaths 0–6 months prior to the census to 30 percent of deaths registered 6–12 months prior (Billings 1885, xlii).

Despite severe under-enumeration, researchers have made creative use of the mortality censuses. By matching deaths registered in the DRA to deaths registered by the mortality censuses, Condran and Crimmins were able to estimate undercounts in both sources and make a more accurate comparison of urban and rural mortality (1980, 188–190). Ferrie used surviving original manuscript returns from the 1850 and 1860 mortality census to link decedents to their household of origin and was thus able to investigate mortality differentials by age, occupation, wealth, nativity, migration status, and household size (2003). The use of linked microdata allowed Ferrie to make another important innovation: by relying only on deaths reported in the six months prior to the census, Ferrie was able to significantly reduce respondent recall error and construct adult life expectancy estimates for white males by region, urban/rural residence, and nativity (1996). The results suggest a substantial advantage in life expectancy at age 20 for white males living in rural areas and for native-born males.

Haines (1998) has made the most significant attempt to use the mortality censuses to construct life tables. He began by observing that the underreporting of deaths for individuals age 5–9, 10–14, and 15–19 appeared to be small. By fitting age-specific mortality rates for these age groups to model life tables, Haines was able to avoid relying on age groups experiencing substantial underreporting of deaths and to construct life tables for the white, black, and total populations by sex for each census year between 1850 and 1900. These tables are clearly superior to their predecessors and a major step forward in our understanding of late nineteenth-century mortality. Despite some concern about regional and temporal differences in undercounting, mortality data were collected for the entire nation. Thus, with the exception of Meech’s 1830–60 life table, Haines’ tables can be considered the only nationally-representative life tables for the nineteenth-century United States. The availability of life tables every ten years between 1850 and 1900 also filled many of the gaps between existing life tables. Contrary to most prior assumptions, Haines’s life tables indicated that mortality did not begin its secular decline until relatively late in the century. Life expectancy at birth was variable without trend between 1850 and 1880—ranging between 38.3 and 44.0 years for both sexes combined. Between 1880 and 1900, however, life expectancy at birth increased from 39.4 to 47.8 years (U. S. Model, both sexes combined).

Researchers relying on the Haines life tables need to be aware of a few potential problems with their interpretation and use. First, as Haines noted, the life tables represent mortality conditions only in the year preceding each decennial census and thus may not be representative of the period or decade in which they nominally represent. Haines’ 1850 life table, for example, like Jacobson’s 1850 life table, may overstate mortality because of the 1849 cholera epidemic. Interpolating between Haines’s life tables for the intercensal periods between 1850 and 1880 suggests that individuals living in the 1860s enjoyed the period’s lowest mortality. The opposite is likely true. During the 1860s the United States suffered four years of civil war, a major and prolonged depression in the postwar South, and, in 1867, another major epidemic of cholera. The war alone is believed to have resulted in the death of approximately 8 percent of white men aged 13 to 43 in 1860 (Vinovskis 1989). Finally, users of the Haines life tables should also be aware that the shape of age-specific mortality rates are strongly influenced by the Haines’ choice of models: model “West” of the Princeton regional model life tables and a “U. S. Model” derived from the 1900–02 DRA life table. As discussed below, there is evidence that these models fail to accurately describe the age profile of mortality in the nineteenth-century United States, particularly for women in their childbearing years. Despite these qualifications, Haines’ life tables are a major point of reference for the latter half of the nineteenth century.5

The only studies of life expectancy prior to 1850 approaching the geographic coverage of the Haines life tables are genealogical-based estimates of adult life expectancy by Kunze (1979), Fogel (1986), and Pope (1992), and mean age at death estimates by Kasakoff and Adams (1995). Because genealogies observe individuals from birth to death, cohort life expectancies are easily calculated. Period estimates can also be made by observing deaths and years of exposure over a given interval, typically a decade. Decennial life expectancy estimates thus reflect mortality over the entire decade, not just a single year. And because individuals are followed over time and space, genealogical data allow the application of event-history methods and more sophisticated analyses. Kasakoff and Adams, for example, were able to examine the impact of migration on subsequent mortality (2000). There are several drawbacks to the use of genealogical data for estimating mortality, however, including substantial under-reporting of infant and childhood deaths (thus limiting estimates to adult life expectancy), under-reporting of female deaths, a bias toward larger and longer-lived families, lack of coverage of the nation’s black and foreign-born populations, small sample sizes for early birth cohorts, a bias toward married individuals who reproduce, and a bias toward families originating in the Northeast and living in the North. Kasakoff and Adams’ dataset, for example, was drawn from nine published genealogies of families whose ancestors settled in seventeenth-century New England. Although nineteenth-century descendents of the nine families can be found in all parts the nation, they were primarily located in the nation’s northern census regions. Kunze and Pope’s datasets were drawn to be more representative of the regional distribution of the United States population. Although not perfectly representative, the geographic coverage of both samples is reasonably representative of the overall population.

Figure 1 plots estimates of white male life expectancy at age 20 by Kunze, Pope, and Haines, and mean age at death estimates for white males known to survive to age 20 by Kasakoff and Adams.6 Four observations can be made. First, the three genealogical studies report very high adult male life expectancies in the late eighteenth and early nineteenth centuries; if the estimates are correct, adult life expectancies in the United States at the turn of the nineteenth century were the highest in the world and were not again exceeded in the United States until circa 1920, approximately four decades after the onset of secular mortality decline. Second, life expectancy estimates by Haines are about three years lower, on average, than those reported in the genealogical studies in the decades where they overlap and can be reliably compared. Third, although there is much variation in each study’s sources, methods, and results, it is nonetheless clear from Figure 1 that the genealogical-based studies support Haines’ contention that mortality did not begin its secular decline until late in the century. Finally, all three genealogy-based studies suggest a significant increase in mortality in the antebellum era, especially in the three decades between 1830 and 1860. White male life expectancy at age 20 was approximately six years lower at mid century than it was in the late eighteenth century.

Male life expectancy at age 20

If correct, a substantial mid-century increase in mortality represents a paradox; based on an assessment of the expected impact of urbanization, public health, and economic growth, Easterlin (1977) had hypothesized a substantial mortality decline before 1880. Although urbanization increased during the period, facilitating the spread of infectious disease and higher mortality, Easterlin noted that the percentage of the United States population living in urban areas remained modest until late in the century. The urban population, for example, was just 28.2 percent in 1880.7 Given an expected 10-year urban-rural differential in life expectancy—an approximate differential suggested by several studies—and assuming a negligible role of public health before 1880, Easterlin estimated that urbanization between 1800 and 1880 reduced life expectancy at birth 2.1 years, all else being equal. The negative effect of urbanization, however, was more than compensated for by increases in the standard of living. Real national income per capita increased dramatically in the period before 1880, leading to significant improvements in diet and housing.8 By assuming a theoretical relationship between life expectancy and per capita income suggested by cross-sectional national data for the twentieth century (Preston 1975), Easterlin estimated that growth in real income in the period 1800–1880 should have increased life expectancy by 14 years. Together with the negative impact of urbanization, Easterlin’s model suggested that life expectancy at birth increased 11.9 years between 1800 and 1880.

Although a reasonable theoretical argument for declining mortality, Easterlin conceded serious doubts in estimates of national income in the period before 1840, the appropriateness of using the relationship between income and life expectancy in the twentieth century to infer the relationship a century earlier, the possibility that public health worsened between 1800 and 1880, and the need for more empirical research. Given these doubts, new estimates documenting a mid-nineteenth century mortality increase cannot be dismissed on theoretical grounds. Moreover, indirect support for an “antebellum paradox” of increasing mortality during a period of strong economic growth is provided by new research on the anthropometric history of the nineteenth-century United States. Fogel first called attention to the positive long-run correlation between cohort life expectancy at age 10 and the final achieved heights of white men. Both series decline in the early to mid nineteenth century and increase late in the century (1986, 464–467). Accumulating evidence from other sources confirms a substantial decline in male height for cohorts born in the mid nineteenth century. Dora L. Costa and Richard H. Steckel, for example, documented a decline in stature among native-born white males from a mean of 173.5 centimeters in the 1830 birth cohort to 169.1 in the 1890 cohort, followed by a substantial and sustained increase in heights for cohorts born in the twentieth century (1997, 72). While identification of the causes of the decline has been difficult—hypotheses include deteriorating diets, a worsening disease environment, the negative impact of early industrialization and urbanization, increasing rates of internal migration, and rising inequality—all researchers have agreed that heights declined significantly. In a recent investigation of the link between antebellum mortality, heights, and net nutrition, Michael R. Haines, Lee A. Craig, and Thomas Weiss (2003) have pointed to the importance of an increasing nationalization and internationalization of the disease environment. Regardless of the ultimate causes, the positive correlation between stature and life expectancy is additional evidence that the decline of life expectancy in the mid nineteenth century reported by recent studies reflects a real increase in mortality.

There are ample reasons to remain skeptical of the overall level of life expectancy reported by the genealogical studies and the size of the suggested decline, however. Genealogical records suffer from two types of bias: a selection bias incurred by selecting data from demographically-successful, native-born families, and a censoring bias incurred by excluding individuals without complete birth and death information from the analysis. Although these biases act in opposite directions—selection bias causes life expectancy estimates to be biased upwards while the censoring bias typically imparts a downwards bias—it is unlikely that they counteract each other perfectly and consistently.9

Adult life expectancy estimates based on genealogical sources tend to be much higher than estimates based on other types of sources, suggesting that selection bias dominates. Between 1785 and 1814, graduates of Yale College—an elite New England population with nearly complete, high quality demographic data—had a life expectancy at age 20 of 40.4 years; Kunze and Pope’s genealogical estimates for the same period are much higher, in the mid to upper forties (Hacker 1996, 121). Adult life expectancies of other elite colonial populations were even lower than that enjoyed by Yale graduates and were especially low in the colonial South. Life expectancy at age 20 was 36.2 years for men graduating from Princeton College between 1709 and 1819; 34.7 years for Maryland legislators born between 1750 and 1764; and 31.7 years for South Carolina legislators born 1750–1764 (Levy 1996; Hacker 1996). Even if we assume no significant socioeconomic status differentials in adult mortality, these studies suggest that genealogical sources overestimate male life expectancy at age 20 at the turn of the nineteenth century by 5–10 years or more. Daniel S. Levy indicates that lower life expectancy in the colonial South was rapidly disappearing by the late eighteenth century, however, suggesting that the overstatement of male life expectancy by genealogical sources was on the lower side of that range, perhaps 6 years in the last decade of the century (1996).

The tendency of genealogical estimates to overstate adult male life expectancy appears to have been lower in the mid and late nineteenth century. In the two periods where they can be compared—1850–60 and 1870–90—Kunze and Popes’ combined estimates of male life expectancy at age 20 are 2.73 years higher, on average, than Haines’ estimates.10 Male life expectancy estimates derived with two-census methods suggest a similar differential. Table 3 shows the results of applying the Samuel Preston and Neil Bennett’s two-census method (1983) to the native-born white population enumerated in the 1850 and 1860 IPUMS censuses (Preston and Bennett 1983). The method assumes the population is closed to migration, a reasonable assumption for the native-born population of the nineteenth-century United States. Although the results may be biased by differential undercounting and the accuracy of age reporting in the two censuses, the resulting life table suggests that genealogical estimates overstate male life expectancy at age 20 in the 1850s by about 3.5 years. Unfortunately, substantial under-enumeration of the 1870 census (see Anderson 1988, 78–82; Steckel 1991) limits comparison to the decade 1850–60.

Table 3

Application of the Preston-Bennett Census-Based Method to the Native-Born White Population, by Sex: 1850–1860

The lower tendency of genealogical sources to overstate life expectancy in the mid and late nineteenth century may be the result of greater migration censoring in the genealogical data. The opening of the trans-Appalachian West with the Treaty of Paris in 1783, the defeat of the Pan-Indian alliance in 1793, land reforms in the early nineteenth century, and the “transportation revolution” of the 1830s likely increased the level and typical distance of internal migration. In the seven decades between 1790 and 1860 the area of the United States increased from 891,364 to 3,021,295 square miles and the number of states from 16 to 33, with the greatest increases between 1840 and 1860 (Anderson 1988, 241, 246). The mean center of population moved further west in the two decades between 1840 and 1860—135.4 miles—than in any other comparable period in United States history (U. S. Bureau of the Census 1921, 34). Although we cannot be sure of the size and timing of the effect—Kunze and Pope do not report the percentage of their study populations with missing death dates by decade—migrants are more likely to be lost from observation. Without adequate attempts to adjust the population at risk, an increase in the percentage of right-censored cases would bias life expectancy estimates downwards, all else being equal.11

Selection bias may also have been less important in the nineteenth century than in the eighteenth century. If selection bias is a function of the propensity of a long-lived ancestor to produce a large number of descendents—thus increasing the odds of producing a future genealogist—the life expectancy of earlier birth cohorts is more critical to the subsequent number of descendants than that of later, larger cohorts, where we can expect more heterogeneity. Put another way, the chances that a couple will produce any descendents beyond a few generations is low if their mortality or the mortality of their children and grandchildren is high. If mortality is low in the first few generations, however, the chances are very high that there will be thousands of descendants (and many potential genealogists) regardless of the level of mortality in subsequent generations (for a general discussion of these issues with regards to Chinese demographic history see Zhao 2001).

Despite concerns about selection and censoring biases, it is clear from recent studies that mortality increased significantly after 1830 and remained relatively high until the 1870s, at which point it began its long and sustained decline. Although genealogical-based estimates of male life expectancy are biased upwards, especially in the eighteenth and early nineteenth centuries, they represent our best source for decennial trends in life expectancy between 1790 and 1890. With care, the estimates can be combined and adjusted to construct a reasonable series of adult life expectancies.

Table 4 attempts such a series by averaging the Kunze and Pope estimates of male life expectancy in each decade and adjusting the combined estimates by a correction factor suggested by comparisons with other studies. Column A shows the average Kunze and Pope estimate of male life expectancy at age 20 for each decade between 1790 and 1890.12 Column B shows a suggested correction factor for each of these decades: -6 years in the late eighteenth century (suggested by comparisons with the graduates of Yale College and other special populations) and -2.73 years in the mid to late nineteenth century (suggested by comparison to Haines’ life tables). The correction factor is interpolated between the 1790s and the 1850s, corresponding to suspected trends in regional migration. The adjusted male life expectancy estimates are shown in column C. Because Pope and Kunze’s genealogical estimates for adult life expectancy end with the 1880–89 decade, the suggested male estimate for the period 1890–99 was obtained by interpolating between the 1880–89 estimate and an estimate obtained from the 1900–02 overall and rural DRA life tables, weighted to reflect the national level of urbanization. (The 1900–02 DRA life tables and their weighting to reflect national levels of urbanization is described in more detail below.)

Table 4

Suggested Best Estimates for Male Life Expectacy at Age 20

Correction factors for the early part of the century are clearly larger and more speculative than those in the second half of the nineteenth century. Indirect evidence suggests that they are approximately correct, however. Given the age structure of the population reported in the United States census of 1800, the adjusted estimates in column C imply a crude birth rate for the white population of 51.5 births per thousand inhabitants. The unadjusted estimate, on the other hand, would imply a crude birth rate of 45.6 per thousand, while a - 2.73-year adjustment would imply a birth rate of 47.7 per thousand. Contemporary observers and twentieth-century demographers have agreed that the birth rate at the turn of the nineteenth century was between 50 and 57 per thousand, strongly suggesting that the 6-year adjustment is justified (Grabil, Kiser, and Whelpton 1958, 5; McClelland and Zeckhauser 1982, 71).13 Although based in part on trends in internal migration and the known impact of migration censoring on mortality estimates, and in part on the observed bias in the genealogical-based estimates of life expectancy compared to other sources, the linear interpolation of the adjustment factor between the 1790s and 1850s is also speculative. As a result, life tables constructed from these estimates will have a larger margin of error than life tables constructed from estimates for the latter part of the century.

The adjusted estimates shown in column C suggest that male life expectancy at age 20 declined approximately three years between 1790–99 and 1850–59. Male life expectancy continued to decline in the 1860s, due largely to the impact of the Civil War. Thereafter, life expectancy began its long-term, sustained increase. It is unlikely that mortality was under significant human control until circa 1880, however. The adjusted series suggests that male life expectancy at age 20 did not exceed its level in the late eighteenth century until the 1880s.

The suggested series indicates a more moderate decline in antebellum life expectancy than the six-year decline suggested by the unadjusted genealogical estimates. The decline is still large, however, and remains a puzzling aspect of nineteenth-century United States demographic history. The suggested revisions shown here do not negate scholars’ characterization of the decline as an “antebellum paradox” or the need for more research on the causes of declining health and longevity during a period of rapid economic growth.

Sex Differentials in Adult Life Expectancy

Estimating female life expectancy at age 20 using genealogical records is a major challenge. Because women appear less often in public records and change their surname at marriage, they disappear from observation more frequently than men. And because genealogical records do not record when right-censored individuals exit observation, female estimates of life expectancy are based on fewer cases and subject to more censoring biases than male estimates.14

Difficulties determining when women entered and exited observation and small sample sizes in each decade likely explain the highly variable sex differentials in adult life expectancy reported by Kunze and Pope (see Table 1). Pope reported that women experienced a 1.6-year advantage in life expectancy at age 20 in the 1820s and a 4.4-year disadvantage in the 1840s. Kunze reported that females had a 3.4-year advantage in the period 1830–34 and a 2.3-year disadvantage in 1835–39. Such rapid shifts in sex differentials in life expectancy are likely spurious and related to poor data quality.

Unfortunately, there are few studies of eighteenth- and early nineteenth-century female life expectancy that can be used to evaluate potential biases in Pope and Kunze’s estimates. Female life expectancy estimates derived using other sources and methods (e.g., estimates from community-based reconstruction studies) are also based on incomplete data and subject to substantial selection and censoring biases (see Hacker 1997,for a summary of existing studies and discussion of potential biases). Life expectancy of women married to Yale graduates at age 20, for example, was 5 years lower in the late eighteenth than the estimates reported by Kunze and Pope. Although the difference is approximately equal to the difference observed between the genealogical estimates and the life expectancy of Yale graduates, more than one-in-four Yale wives had an unknown date of death, rendering an assessment of bias in the genealogical estimates uncertain. Given different assumptions about the mortality experiences of women with a missing death date, the life expectancy of Yale wives at age 20 may have been one year higher or lower (Hacker 1996, 83, 98). Much higher proportions of missing data and margins of error characterize other late eighteenth- and early nineteenth-century estimates of female life expectancy.

For the late nineteenth century, Kunze and Pope’s estimates of the female life expectancy can be compared with Haines’ estimates. The comparison indicates that Kunze and Pope’s combined estimates for white females at age 20 in are slightly lower (-0.36 years) than Haines’ estimates in the years in which they can be reliably compared. This is in sharp contrast to the comparison with Haines’ estimate for white males, where the genealogical-based estimates were substantially higher (2.73 years at age 20). Given the high proportion of missing death records for women in genealogies, the difficulties determining when women entered and existed the at risk population, and the highly variable sex differentials in life expectancy reported by Kunze and Pope, it is tempting to conclude that this discrepancy is due entirely to bias in estimating female life expectancy from genealogical data. Some portion of the difference in the male and female comparisons with Haines’ life table estimates may be due to Haines’ choice of a model life table system, however. This possibility is explored in the subsequent section examining age patterns in nineteenth-century mortality.

Regardless of the ultimate cause, poor data quality, inconsistent results, and the lack of an independent assessment of potential bias strongly suggests that determination of the level of and trend in female life expectancy is best inferred from male estimates. This section discusses sex differentials in nineteenth-century life expectancy, suggests a best estimate for the differential at age 20 in each decade, and calculates the resulting series of female life expectancy from the adjusted male estimates shown in Table 4. The sex differential is assumed to be constant before 1860, after which fertility and mortality decline are assumed to have contributed to more rapid female gains in life expectancy relative to male gains (see Preston 1976, chapter 6, for a discussion of the impact of mortality decline on sex differentials in mortality). Estimates are made separately for the 1860s to account for excess male mortality during the Civil War.

The best estimate of the sex differential in life expectancy for the period before 1860 and the best estimate for each decade after 1870 are not obvious from existing studies of nineteenth-century U.S. mortality. Kunze and Pope’s estimates suggest a male advantage in life expectancy at age 20 while Haines’ life tables suggest a female advantage. On average, the combined Pope and Kunze estimates of male and female life expectancies at age 20 indicate a 0.9-year male advantage before 1860. For census years 1850 and 1860, Haines’s U. S. Model life tables suggest an average female advantage in life expectancy at age 20 of 1.1 years (Haines’ life tables based on Princeton model West life tables indicate a 2.9-year female advantage).

These contrasting results persist in the postwar era. Kunze and Pope’s results indicate that males enjoyed a 2.4-year advantage, on average, in the 1870s and 1880s while Haines’s U. S. Model life tables indicate a 1.3-year female advantage (2.7 years using model West). At the beginning of the twentieth century, the 1900–02 DRA life table shows a 1.6-year female advantage in life expectancy at age 20, which is in close agreement with Haines’ U. S. model. The close agreement is not surprising, of course; Haines’ U. S. model life tables are based on the age-pattern of mortality in the 1900–02 DRA. The life table constructed for the rural parts of the 1900–02 DRA, however, shows a female advantage in life expectancy at age 20 of just 0.1 years, closer to the implied sex differential in the combined Pope and Kunze estimates.

The different sex differentials in adult life expectancy observed in the overall and rural DRA life tables hint that males may have enjoyed higher adult life expectancies in the more rural past. Such a conclusion is supported by the demographic literature on nineteenth-century European populations.15 A recent comparative study of mortality in rural villages in eighteenth and nineteenth-century Europe and Asia (the Eurasia Population and Family History Project), for example, reports lower female lower life expectancy at age 25 in three of the four European study areas. Sex differentials in life expectancy at age 25 was -2.3-years for Sart, Belgium (a 2.3-year female disadvantage relative to males); -1.0 years for Casaluidi, Italy; -2.8 years for Madregolo, Italy; and 0.7 years for Scanian parishes in Sweden, for an unweighted average of -1.4 years (Campbell et al. 2004, 66). Lower female life expectancy at age 25 resulted from a remarkably consistent pattern of higher female mortality during prime childbearing ages across study populations, suggesting that maternal mortality and maternal depletion played a large role in the consistent pattern (Alter et al. 2004). The pattern is characteristic of mortality in national populations with life expectancy below 45 and suggestive of higher female mortality from pulmonary tuberculosis, other infectious diseases, and maternal causes (Preston 1976, 91).

Some evidence suggests that females in rural areas of nineteenth-century Europe suffered higher rates of infectious disease relative to males than females in urban areas. Dominique Tabutin and Michel Willems, for example, cite evidence that excess female mortality and susceptibility to respiratory diseases such as tuberculosis were more pronounced in rural areas (cited in Alter et al. 2004). Excess female mortality extended over a greater range of ages and was much higher in England’s 63 “healthy districts”—mostly rural districts with crude death rates below 17 per thousand—than in the 1838–54 English Life Table (Woods 2000, 187). According to Shelia Ryan Johansson, a probable reason for the higher incidence of tuberculosis among females and higher rates of female mortality in rural areas of Victorian England was lower nutritional status. Agricultural societies in the past, she observed, routinely discriminated against females by reserving most food and the vast majority of meat for husbands and sons. Industrialization and the ability of women to participate in the paid labor force eventually ended this nutritional discrimination (Johansson 1977). Higher fertility is another possible reason for higher female mortality in rural areas. Although maternal mortality rates were low relative to mortality rates from tuberculosis—most nineteenth-century estimates suggest that maternal mortality averaged between 5 and 10 maternal deaths per thousand live births (Kippen 2005)—higher rates of nuptiality and martial fertility in rural areas increased the cumulative risk of maternal mortality. Perhaps more importantly, pregnancy and lactation imposed greater nutritional demands on women and reduced cell-mediated immunities, increasing the risk of contracting tuberculosis and other opportunistic infections.16

Unfortunately, with the exception of Kunze and Pope’s studies, estimates of sex differentials in life expectancy for the nineteenth-century United States are based on highly urban, low fertility populations such as Massachusetts in the late nineteenth century, the 1900–02 DRA, or else, like Haines’ life tables, are derived from models based on these populations. The 1850–60 Preston-Bennett life table (Table 3 above), however, avoids this urban, low fertility bias by relying on the national native-born white female population in the 1850 and 1860 IPUMS samples. The results suggest sex differentials in life expectancy similar to Kunze and Pope’s genealogical-based estimates. At age 15, the sex differential in life expectancy was -1.2 years, rising to -2.3 years at age 20. The male advantage in life expectancy lasted until age 35. Thereafter, females enjoyed a slight advantage in expected remaining years of life.

Together, the results from eighteenth- and nineteenth-century European populations and the results indicated by the 1850–60 Preston-Bennett life tables for native-born whites suggest that the overall average 0.9-year male advantage in life expectancy at age 20 reported by Kunze and Pope for the period 1780–1859 was approximately correct.17 As indicated by the 1900–02 DRA life tables, however, a female advantage in life expectancy at age 20 had emerged by the turn of the twentieth century. If the overall and rural 1900–02 life tables are weighted and combined to approximate the urban percentage of the national population, the female advantage in life expectancy at age 20 was 0.9 years in 1900.18

Table 5 suggests best estimates of female life expectancy at age 20 between 1780 and 1860 by assuming a fixed 0.9-year advantage in male life expectancy. As shown in column B, the sex mortality differential was assumed to shift in favor of females in a linear fashion between the 0.9 female disadvantage in life expectancy in the period before 1870 and the 0.9-year advantage in female life expectancy suggested by the weighted 1900–02 DRA life tables. Although somewhat speculative, the linear shift from a male advantage to a female advantage in life expectancy between 1870 and 1900–02 is consistent with known changes in sex mortality differentials accompanying mortality decline, the epidemiological transition, and fertility decline. The decline in pulmonary tuberculosis, in particular, likely led to more rapid declines in female mortality relative to male mortality (Preston 1976, chapter 6). Because excess male mortality during the Civil War likely affected sex differentials in mortality, the female estimate of life expectancy in the period 1860–69 was obtained by averaging the adjusted female life expectancy in the 1850s and 1870s. Suggested best estimates of female life expectancy at age 20 are shown in column C.

Table 5

Suggested Best Estimates for Female Life Expectacy at Age 20

The Age Profile of Nineteenth-Century Mortality

Mortality varies with age in a consistent pattern, sometimes characterized as a “U” or “J” shape, across a wide range of mortality levels. Mortality rates are very high in infancy, drop rapidly in childhood, reach their lowest level in late childhood and adolescence, and then begin to increase in a fairly regular manner with age. Because of this consistency, demographers have long sought to model mortality as a function of age and overall mortality. Among other uses, an accurate model would make it possible to identify deviations in empirical data from model patterns (suggestive of particular conditions or poor data quality), to gain insight into environmental and behavioral factors that may determine deviations, and to construct life tables from poor data, partial data or even a single parameter (Preston, Heuveline and Guillot 2001, 191–192). With an accurate model, for example, it would be possible to generate decennial life tables from the estimates of adult life expectancy suggested in Tables 4 and ​5. Choice of model, however, involves some guesswork and is a potential source of substantial error.

Three basic approaches have been used to model the age pattern of mortality: mathematical approaches that represent mortality as a function of age, tabular approaches that show expected patterns of age-specific mortality rates and other life table parameters at different mortality levels, and a combination of the first two approaches that uses a mathematical function to relate mortality in a given population to a tabulated standard population (Preston, Heuveline and Guillot 2001, 192–201). Early attempts to describe the relationship between mortality and age with a single mortality function were unsuccessful (see Woods 2000, 170–190 for a discussion of nineteenth-century attempts to specify the “laws of vitality”). For a variety of reasons, including changes in behaviors and in the leading causes of death (e.g., smoking and cancer), the age pattern of mortality varies enough across time and space that a simple mathematical model is not practical. An attempt by Heligman and Pollard (1980), for example, required a complex equation with eight parameters to model the age profile of mortality from infancy to old age.

The second approach to modeling age patterns of mortality has been the publication of model life table systems—sets of “model” life tables at different levels of morality. The most popular set of model life tables, the Princeton regional models, were published by Princeton demographers Ansley J. Coale and Paul Demeny in 1966 and revised in 1983 (1983). Coale and Demeny examined empirical data from 326 historical and contemporary populations. From the 192 life tables deemed reliable, Coale and Demeny identified four regional patterns, which they used to construct four “families” of model life tables. In the 1983 revision male and female life tables are shown at 25 different levels of mortality, ranging from Level 1 (female life expectancy at birth equal to 20 years) to Level 25 (female life expectancy at birth equal to 80 years) for each of the four regional patterns. Intermediate levels are easily obtained by interpolation. The four groups closely conform to four regions of Europe, which was the primary source of the life tables. The “North” model is based largely on life tables from Scandinavian countries. It is characterized by low infant mortality and low mortality at older ages. The “East” model is based on life tables from Eastern Europe and is characterized by high infant mortality. The “South” model is based mostly on tables from Southern Europe, and is characterized by high mortality under age 5 and above age 65 and low mortality between age 40 and 60. The “West” model is more of a residual group and is based on the largest number of life tables, including tables from Western Europe, the United States, Canada, Australia, New Zealand, and Japan. Other model life table systems—including those created by the United Nations—have been created for developing countries in Asia, Africa, and Latin America, where different environments and causes of death lead to different patterns of mortality than are found in Coale and Demeny’s European-dominated system (United Nations 1982).

For populations with poor vital registration data, the choice of a model life table—and thus the assumed age profile of mortality—typically requires some guesswork. Colin Newell notes that the “general, but not always helpful, rule is to use a [model life table] system which is flexible enough to let real features and irregularities through, but which is sufficiently robust to be unaffected by errors in the data” (1988, 165). Because the age profile of mortality is largely the result of environmental and behavioral factors—which determine the distribution of causes of death and the level of mortality—most analysts try to rely on a model life table system based on data from a nearby region with a similar environment, behaviors, and level of mortality. U.S historical demographers tend to rely on Coale and Demeny’s model West, which is based in part on historical life tables for the United States (including the 1900–02 DRA life table). Robert V. Wells, for example, used model West to infer life expectancy at birth in colonial America from adult and child survival estimates reported in various studies (1992). Suspecting probable under-enumeration of infant deaths in the 1900–02 DRA, Condran and Crimmins fitted mortality rates for the 1 to 4 age group to model West life tables in order to estimate life expectancy in urban and rural areas of the DRA (1980, 191). Where it can be compared to empirical data, model West appears to be a good fit for the total and white populations of the early twentieth-century United States (Haines 1979, 197; Preston and Haines 1991, 66). Douglas Ewbank, however, found that the age mortality profile of early twentieth-century black population of the United States more closely matched the United Nation’s “Far East” model life table (1987).

Depending on the application, the choice of model can be important. Preston and Haines found that choice of regional model had very little impact on indirect estimates of child mortality in the 1900 census (1991, 64–67). Estimating infant mortality and life expectancy at birth from life expectancy at age 20, however, is problematic. Table 6 shows implied estimates of male and female life expectancy at birth, infant mortality rates, and the proportion of the population surviving to age 20 when male and female life expectancy is 40 years using the four Princeton regional models. The implied life expectancy at birth for males ranges from a high of 43.8 years in model West to a low of 38.0 years in model East, a difference of nearly 6 years. Implied male infant mortality rates vary from a low of 156 per thousand in model North to a high of 250 in model East. Using model West, nearly 72 percent of the population survived to age 20. In model East the percentage was less than 62 percent. Similar differences are observable for the female population. These differences illustrate the large potential error that can be incurred by relying on the wrong model to infer a complete life table from a single parameter.

Table 6

Implied Coale and Demeny life table parameters when e20=40 years

The third approach to modeling the age profile of mortality, developed by William Brass (1971), uses a mathematical function to transform a standard life table. It thus represents a combination of the mathematical and tabular approaches. Brass observed that logits of the lxs from any two life tables are related to one another by a linear relationship, making it possible to describe a set of logits in an observed or target population using the logits from the standard table and appropriate intercept and slope values. Briefly, the logit transformation of the lx column is based on the equation

logit(1 − lx) = Yx = 0.5 Log e(1 − lx/lx) , 

(1)

where l0 = 1.0. The logits of an observed population, YObs(x), are related to the logits of a standard population, Ys(x), by the linear equation

logit (Obs. lx) = YObs(x) = α + β Ys(x).

(2)

To fit an observed life table to a standard table, logits of the observed lxs are plotted against the logits of the standard life table. A straight line is then fitted to the points (typically with simple linear regression or weighted regression techniques), and the intercept and slope of the line, α and β, are calculated. Once α and β are calculated, fitted logits can be computed from the standard logits, and the anti-logits can be taken to produce a set of fitted lxs, as shown in the equation below:

Fittedlx=11+e2YFit(x).

(3)

When the intercept (α) equals 0 and the slope (β ) equals 1, the standard table will be reproduced. Values of the intercept parameter greater than 0.0 will shift the level of mortality above the standard table and values less than 0.0 will shift the level of mortality below the standard table. The slope parameter determines the “tilt” of the table. A slope value greater than 1.0 indicates that infant and child mortality is lower relative to adult mortality than in the standard table, and a slope less than 1.0 indicates that infant and child mortality is higher relative to adult mortality than in the standard. It thus becomes possible to construct a family of related life tables from a standard life table by varying the intercept and slope parameters, calculating the anti-logits of the resulting values, and constructing the resulting life tables.

Although Brass suggested two sets of logits to use as a standard—a general standard and an African standard—any life table can be used and logits calculated directly from the lx column. Appropriate choice of a standard table can preserve variations in the age profile of mortality that cannot be obtained by varying the slope and intercept parameters of a standard table, such as the level of older age mortality relative to mid age mortality or the level of infant mortality relative to childhood mortality.19 To construct his “U. S. Model” life tables, for example, Haines relied on the 1900–02 DRA life table as a standard table. With the help of available historical life tables from Massachusetts and other United States life tables of reasonable quality, Haines first estimated the impact of urbanization and time on the slope of the age mortality profile. While more urban environments increased infant and childhood mortality relative to adult mortality, the trend in the late nineteenth century was toward relatively lower levels of infant and child mortality (Haines 1979, 303). From this relationship Haines determined the likely slope parameter in each census year between 1850 and 1900, effectively reducing the two-parameter logit model to a one-parameter model. The final intercept parameter was determined by fitting the age-specific death rates of children age 5–19 in the mortality censuses (Haines 1979).

Comparison of Haines’ U. S. Model life tables with the life tables constructed using model West as a standard indicates that the U. S. model typically yields higher infant mortality rates, lower adult mortality rates, and lower life expectancy estimates at birth. In 1880, for example, Haines’ U. S. model suggests an infant mortality rate of 0.214 for white males and a life expectancy at birth of 40.4 years. The life table constructed using model West suggests an infant mortality rate of 0.180 and a life expectancy at birth of 40.9 years.

Arguably, the 1900–02 DRA life table is a more appropriate standard for the nineteenth-century United States than a generic standard or even model West.20 As noted in Table 2 above, however, the DRA population was much more urban than the overall population in 1900, had a higher proportion of the population foreign born, a lower proportion engaged in agriculture, and much lower fertility. The contrast is even greater with the overall population in the early and mid nineteenth century United States, which was overwhelmingly rural and had very high fertility. Although variation of the slope parameter can pick up some of the suspected impact of urbanization and time on the suspected age profile of mortality in the nineteenth century, the increase in urbanization and immigration, the decline in fertility and the agricultural sector of the economy, and the onset of the public health movement and epidemiological transition in the later part of the nineteenth century likely affected the distribution of causes of death and the age profile of mortality in more complex ways. It is likely, for example, that declining tuberculosis in the late nineteenth century had a significant impact on the mortality of young adults relative to infants and older adults, especially among females. Condran and Cheney report that the decline in mortality from pulmonary tuberculosis explained 26.8 percent of the decline in mortality in Philadelphia between 1870 and 1900 and was overwhelmingly important in the decline in death rates at ages 20–39 (1982, 105).

In addition to mortality decline, rapid fertility decline in the late nineteenth century (Hacker 2003) likely had an impact on the age-specific mortality rates of females. Although maternal mortality rates were lower than typically imagined in the qualitative literature (Schofield 1986), repeated exposure to death in childbirth in high fertility populations increased female mortality relative to male mortality during childbearing ages.21 Pregnancy may have been a significant risk factor in contracting tuberculosis, the leading killer of nineteenth-century Americans, and other opportunistic infections. We can thus expect that the shape of age-specific mortality rates for females in the early to mid nineteenth century varied significantly from the shape of age-specific rates for females in the 1900–02 DRA, even if the slope of the age profile is adjusted to account for suspected higher infant and child mortality relative to adult mortality prior to the onset of mortality decline.

Some indication of the possible bias can be seen in Figure 2, which compares the age-specific mortality rates for white females in the 1900–02 rural DRA life table with white females in the 1900–02 overall DRA life table. Age-specific rates for white females in the rural DRA were noticeably lower than that for that for white females in the overall DRA at most ages, reflecting the overall higher life expectancy for females in rural areas. Mortality rates were roughly equal at ages 10–14, 20–24, and 25–29, however, and higher for rural females at age 15–19.22 Although we cannot be sure of the causes, higher mortality in rural areas during adolescence and early adulthood are suggestive of higher death rates from tuberculosis and maternal mortality (Preston 1976; Henry 1989). Females age 20–49 residing in rural areas of the DRA had 9.4 percent more own children in the household than females in the overall DRA, increasing their exposure to maternal mortality and risk of contracting tuberculosis and other infectious diseases.

Proportion dying by age group, white females in 1901 Death registration area

Among the four Princeton regional models, age-specific death rates for white males and females in the rural and overall 1900–02 DRA had the closest correspondence with model West (after age 20, males in rural areas of the DRA had a closer relationship with model North). Relative to the model West level corresponding to the same life expectancy at birth, however, female death rates in the 1900–02 DRA and rural areas of the 1900–02 DRA were much higher in peak childbearing years. The difference, as shown in Figure 3, was especially pronounced for females residing in rural areas. With the exception of age groups between 15 and 35, there is remarkably close correspondence between model West level 15.17 and the mortality of women in the rural DRA. Age-specific death rates for rural females between ages 15 and 29, however, exceeded the level expected in Model West by approximately 27 percent. The greatest divergence from the model pattern, 36 percent, was at ages 20–24. Although a similar pattern exists for males (not shown)—higher death rates at ages 5–34 for white males residing in rural areas of the 1900–02 DRA relative to the corresponding model West level, lower rates at ages 40 and above—the differences were much smaller.

Proportion dying by age group, white females in rural areas of the 1901–02 DRA compared to Princeton model West and model North

Similar “humps” in age-specific mortality rates for females between the approximate ages of 15 and 45 have been observed in other historical populations, including eighteenth and early nineteenth-century American populations (Rutman and Rutman 1976; Logue 1991; Hacker 1996), the mostly rural eighteenth- and nineteenth-century populations studied by the Eurasia project (Alter et al. 2004), and the mid nineteenth-century population of England (Wrigley and Schofield 1989 [1981], 708–709). In their reconstruction of English population history, for example, E. A. Wrigley and R. S. Schofield noted that while age-specific mortality rates of males in the third English life table (1838–54) corresponded well with model North of the Princeton regional life tables, females had higher than expected rates from age 10 through age 35. The deviation from the model pattern prompted Wrigley and Schofield to construct their own model, based in part of the English life table and in part on model North.

Did the same distinctive hump shape during childbearing years that characterized age-specific mortality rates for females in rural areas of the 1900–02 DRA and various European and Asian populations also characterize the overall population of the nineteenth-century United States? Figure 4 shows the implied proportion dying in each age group from the Preston-Bennett 1850–60 life table shown in Table 3 above and the model West level corresponding to the equivalent life expectancy at age 10. Although the age-pattern of mortality suggested by the Preston-Bennett life table is somewhat erratic, the distinctive deviation in age-specific mortality rates from the expected pattern is again evident. For females, the implied proportion dying in prime childbearing age groups 25–29 and 30–34 exceeded the implied proportion dying in age groups 35–39, 40–44, and 45–49. Although much less pronounced, a hump is also evident in the age-specific mortality pattern for white males. The two age profiles suggest the known age and sex profiles of tuberculosis mortality. The less pronounced hump for males may also indicate the absence of maternal mortality or different patterns of census coverage errors by age. Whatever the ultimate cause, the results of the Preston-Bennett life table suggest that the age-sex-pattern of mortality in the nineteenth-century United States more closely resembled the pattern in the rural areas of the 1900–02 DRA than the pattern in the overall DRA.

Proportion dying by age group, native-born whites in Preston-Bennett 1850–60 life table

Another way of approaching the question is through examination of the sex mortality ratios by age. Despite higher life expectancies in the rural 1900–02 DRA than in the overall DRA, the ratio of male to female mortality was lower at most ages in the rural DRA. The difference was especially pronounced during childbearing ages.23 White females in the rural DRA experienced excess mortality relative to males between age groups 15–19 and 40–44. In contrast, females in the overall DRA experienced lower mortality than males in all age groups. Among the nineteenth-century studies reporting lower female life expectancies in early adulthood cited above, most show excess females mortality relative to males in prime childbearing years. Alter, Manfredini, and Nystedt, for example, report excess female mortality from age 25 to 50 in six of the seven study populations in Sweden, Belgium, Italy, China, and Japan. In the rural village of Sart, Belgium, to cite a typical example, the ratio of male to female probability of dying in the interval 25–50 was 0.78 (2004, 334). England’s third life table (1838–54) shows excess female mortality in all five-year age groups between age 10 and 40 (Wrigley and Schofield 1989 [1981], 709), although the female disadvantage was modest. The lowest male to female mortality ratio, 0.95, was for the 25–29 age group. Excess female mortality was much higher in England’s “healthy districts” (Woods 2000, 187), however, echoing the similar contrast between sex mortality ratios in the rural and overall 1900–02 DRAs of the United States.

Although we lack death-registration data for the nineteenth-century United States, the 1860–1900 censuses of mortality allow the construction of sex differentials by age. Condran and Crimmins’ analysis of these data indicated that while the mortality censuses undercounted infant and elderly deaths, the relative undercount of males and females varied little by age (1979). Figure 5 shows the average sex ratio in mortality in the 1860–1880 censuses by age compared to the ratios in the overall 1900–02 DRA and the rural areas of the 1900–02 DRA. Figure 5 also includes a plot of the average sex mortality ratios in Haines’ 1850–1880 “U. S. Model” life tables. Sex mortality ratios indicated by the census data suggest a similar pattern to the 1900–02 rural DRA pattern: excess female mortality from adolescence through prime childbearing years and excess male mortality at other ages. Sex mortality ratios in Haines’ life tables, however, more closely conform to the 1900–02 overall DRA. Although Haines’ tables indicate modest excess female mortality in childhood and approximately equal sex ratios during prime childbearing years 20–34, the age pattern of sex mortality ratios is much closer to the overall 1900–02 DRA pattern than to the rural DRA pattern. Haines’ tables also suggest a lower sex differential in mortality in infancy than either the 1900–02 overall or 1900–02 rural DRA life tables.

Ratio of male to female probability of dying by age (qx)

Figures 4 and ​5 strongly suggest that the 1900–02 rural DRA life table is more representative of the shape of mortality in the nineteenth-century United States than the overall DRA life table. Age-specific mortality rates implied by the Preston-Bennett 1850–60 life table and sex mortality ratios by age in the 1860–80 censuses of mortality more closely conform to the pattern in the 1900–02 rural DRA life table than the overall DRA life table (which was itself a closer match than model West). The correspondence should not be surprising: like the rural DRA life table, the population of the nineteenth-century United States was less urban, more agricultural, and had higher fertility than the population of the 1900–02 DRA and populations used in the construction of model West. Although we cannot be certain of the true shape of age-specific mortality rates in the nineteenth century, the available evidence indicates that any model used to construct nineteenth-century life tables, especially life tables for the earlier part of the century, should draw more heavily from the 1900–02 rural DRA life table than from the overall DRA life table.

New Decennial Life Tables, 1790–1910

Two life tables constructed by James Glover for the 1900–02 Death Registration Area (DRA) are essential for this project: (1) the life table for the white population residing in the ten DRA states and the District of Columbia, and (2) the life table for the white population in the rural areas of the DRA. When the life tables were published in 1921 the Census Bureau’s definition of “urban” was considered cities of 8,000 or more inhabitants. All other places were considered rural. The Census Bureau subsequently redefined urban as places of 2,500 or more inhabitants. So although nominally non-urban, the 1900–02 “rural” DRA life table is based in part on a population residing in the modern definition of an urban area, albeit modest towns and cities of 2500 to 8000 inhabitants.

As shown in table 2 above, the population living in the 1900–02 DRA was predominately urban: over 60 percent lived in the modern definition of an urban area. Over 13 percent of the population in the rural areas of the DRA also lived in an urban area. The DRA covered 26.2 percent of the national 1900 population; the rural parts of the DRA only 12.0 percent.

What can be inferred about the level and pattern of national mortality in 1900–02 given that nearly three-quarters of the population lived in states that were not part of the DRA? Although we could assume that the larger, more inclusive life table for the overall DRA is more representative of the national population, we know that urbanization, industrialization, nativity, and fertility in the DRA were not representative of the national population and likely affected the shape, level, and sex differential in mortality. A better choice might be the 1900–02 rural DRA life table. Although a subset of the overall DRA, the rural population was more representative of the national population in terms of fertility, nativity, and occupation structure. Unsurprisingly, however, urbanization was higher in the nation as a whole than in the rural areas of the DRA and was likely the most important factor influencing mortality.

The simplest and most defensible inference is to combine the overall and rural DRA life tables, using appropriate weights to produce a life table reflecting the rate of urbanization in the nation as a whole. If we assume that that national population in 1901 was 40.2 percent-urban (an interpolation of the Census Bureau’s estimate of urbanization in the nation as a whole in 1900 and 1910), it is a simple matter to calculate the weight needed for each DRA life table and to combine the two to produce one life table representative of the nation’s urban population.24 Relative to the overall DRA life table, the resulting combined life table would increase estimates of white life expectancy at age 20 by 1.5 years for white males and 0.9 years for white females. Sex differentials in life expectancy at age 20 would fall from a 1.6-year female advantage in the overall DRA life table to a 0.9-year female advantage in the combined table.

The combined table could in turn be used as a model for earlier years: logits of the table’s lx values could be taken and new life tables generated by varying the slope and intercept shown in equation 2 above to construct a predicted set of logits, calculating the lx values by taking the anti-logits using equation 3, and constructing a new life table from the predicted lx values.

There are several problems in such an approach. Most obviously, urbanization was increasing rapidly in the decades before the 1900 census. By design, the combined 1900–02 life table is representative of urbanization in the 1900–02 national population; nineteenth-century populations were far more rural. Haines’ method (1979) is one possible way around this problem. Drawing on his analysis of available late nineteenth-century city and state life tables, Haines observed that the slope of age-specific mortality rates varied across time and by level of urbanization in a predictable way. Haines was thus able to set the slope of his model as needed to fit the period and level of urbanization.

Although a useful innovation, Haines’ method cannot be applied uncritically to decades early in the nineteenth century. Most of the change observed in the slope of mortality likely reflected the impact of public health initiatives between 1880 and 1900 in the nation’s largest cities, particularly efforts to clean water and milk supplies.25 The net result was falling infant and early childhood mortality relative to adult mortality in large urban areas, despite rapidly increasing urbanization. Because most small cities made only modest attempts at public health initiatives before 1900 (Duffy 1990, chapter 12), it is much less certain if infant and childhood mortality fell relative to adult mortality for the nation as a whole between 1850 and 1900, which Haines’ model predicts. Indeed, as Haines noted, Princeton model West suggests the opposite. Between levels 9 and 13—equivalent to an increase in female life expectancy from 40 to 50 years and roughly spanning the increase in life expectancy in the late nineteenth century United States—model West suggests that infant and childhood mortality should increase relative to adult mortality. Only at mortality levels above level 13 does infant and childhood mortality begin to decline faster than adult mortality (Haines 1979, 300–301).

Given this uncertainty, a better approach would be to create a unique standard for each decade of the nineteenth century by repeating the weighting exercise of the 1900–02 DRA and 1900–02 rural DRA life tables described above, using the appropriate weights to yield a new standard life table representative of the urbanization level in each decade. Table 7 shows the results of that exercise. Included in the table are estimates of the mid-census level of urbanization in each decade (an average of the percentage urban in each of the beginning and ending censuses), the corresponding proportional weights of the 1900–02 overall and rural DRA life tables used to create each standard, and the resulting logits of the tables’ lx values by age and sex. Before 1850, the national level of urbanization was below that estimated in the rural 1900–02 DRA table. It was therefore assumed that the rural 1900–02 table represented the standard mortality pattern for all decades before to 1850. After 1850 urbanization began to exceed the level of urbanization in the 1900–02 rural DRA life table, requiring increasing weight to be given to the overall DRA life table. The applied weighting of the 1900–02 overall DRA life table increased from 0.09 in the 1850–59 decade to 0.51 in the 1890–99 decade.

Table 7

Standard life table logit values, Ys(x), for decennial life tables

From there it was a simple matter of varying the intercept in equation 2 above and constructing a new life table to fit the estimates of adult life expectancy shown in Tables 4 and ​5. With one exception, the resulting life tables are shown in Table 8. The exception is the 1860–69 life table for white males, which was modified to account for high mortality among males of military age during the Civil War. It was constructed in three steps. First, a “base” life table for the 1860–69 period was constructed by using the average of the 1850–59 and 1870–79 estimates of male life expectancy at age 20. Second, an estimate of excess male deaths in the 1860–1870 intercensal period was made by cohort using two-census survival methods.26 Finally, the excess male deaths were added to the base life table Table 9 shows the results for each year of the war.27 Unsurprisingly, mortality was highest in 1864, the last full year of the conflict. The estimates imply a white male life expectancy at birth of 26.9 years, likely the lowest level in U.S. history. Although based on crude estimates, the method retains the unusual risk of early death among young white males in the war. The resulting life table for the 1860s suggests a male life expectancy at age 20 of 35.6 years, approximately 2 years lower than the adjusted Kunze and Pope estimate. Although the base life table and the number of excess male deaths could be adjusted to yield a perfect match, it is unclear which estimate to adjust. It is also possible that the genealogical samples, which are known to under-represent individuals who do not marry or reproduce, are biased against soldiers participating and dying in the war. It was therefore decided to make no further adjustments to the life table.

Table 8

New Life Tables for the White Population of the United States, 1780–1900

Table 9

Proportion dying in age interval x to x+n, white males