trends and determinants of adult mortality in early … and...trends and determinants of adult...

J. David Hacker

Trends and

Determinants of

Adult Mortality in

Early New England

Reconciling Old and New Evidence

from the Long Eighteenth Century

Despite decades of research, demographic historians are still uncertain about

mortality trends and determinants in early New England. Although researchers agree that New England mortality was low relative to other re-

gions of early America in the seventeenth century, they disagree about the direction of mortality trends over the course of the eighteenth cen-

tury. Community-based reconstitution studies conducted in the 1960s and 1970s at first seemed to provide strong evidence of a decline in adult life

Social Science History 21:4 (winter 1997). Copyright ? 1997 by the Social Science History Association.

482 Social Science History

expectancy (Greven 1970; Norton 1971; Vinovskis 1972; D. S. Smith 1973).1 More recent studies, however, contradict the assessment of deteriorating health conditions in eighteenth-century New England. Family-based genealogical studies document a substantial long-term increase in adult male life

expectancy over the course of the century, although the increase was followed by a slight decline in the early part of the nineteenth century (Fogel 1986; Kasakoff and Adams 1995).

The limited investigations of mortality determinants are characterized

by similar discrepancies. Reported sex mortality differentials vary widely. At one extreme, a study of seventeenth-century Plymouth concluded that women suffered a seven-year disadvantage in life expectancy at age 20

(Demos 1965). At the other extreme, a study of eighteenth-century Nan- tucket reports that women enjoyed a five-year advantage (Logue 1991).2 The existence and size of geographic mortality gradients before the nineteenth

century is also uncertain. Mortality is believed to have been higher in urban areas than in rural communities, but comparisons are limited to crude death rates and remain inconclusive. No study has estimated class, race, marital status, or occupational differentials in life expectancy, and no study has

reported the statistical significance of its results. Given the uncertainty associated with the data, it should come as no

surprise that historians disagree about the causes and timing of early Ameri- can mortality change. A few researchers argue that the rapid increase in New England's population was a critical factor in facilitating the spread of infectious disease (Norton 1971; Meindl and Swedlund 1977; Dobson 1989; Logue 1991). Diphtheria, smallpox, and yellow fever epidemics in the eighteenth century are well documented (Caulfield 1938; Duffy 1953), although smallpox mortality declined in Boston after the mid-eighteenth century, perhaps as a result of inoculation (Blake 1959). Other researchers stress that

rising levels of income and nutrition in the eighteenth century were largely responsible for increasing life expectancies (Fogel 1983, 1986). Still others contend that medical knowledge, personal hygiene, public health initiatives, and various social and cultural factors played important roles in determining the level and trend of early American mortality.

Without consensus on early American mortality change, speculation on its sources is premature. Laying aside for the moment the admittedly interesting question of causation, this article examines several possible ex-

Trends and Determinants of Adult Mortality in Early New England 483

planations for the discrepancies in the early American mortality literature.

First, it considers the possibility that the different mortality trends reported

by the various studies reflects a spectrum of New England demographic experiences, part of what Richard Archer terms a "New England mosaic"

(1990). Because most investigations have concentrated on small communi-

ties, it is possible that intraregional variations in mortality account for the different results. Various determinants of mortality-a community's population, proximity to the coast, economy, and climate--may have engendered a wide variety of mortality levels and trends in New England. Second, it evaluates possible biases in the published studies. To some degree, all estimates of early American life expectancy suffer from small sample sizes and

incomplete vital records. The treatment of missing data, or "censoring," is considered in some detail. Recent methodological investigations conclude that "migration censoring" -the loss of individuals from observation due to their migration from the area of study--may severely bias life expectancy estimates made by community-based reconstitution studies.

Data

To test the preceding hypotheses, this study relies on a neglected source: bio-

graphical sketches of the graduates of Yale College published by Franklin B. Dexter in the late nineteenth and early twentieth centuries (Dexter 1885-

1912). Dexter, a librarian and historian at Yale, spent most of his life working on the biographies and appears to have examined every potential source. In addition to the records available at Yale, most notably the Alumni Tri- ennial Catalogue and annual obituary records, he consulted genealogies, town records, and tombstones and even corresponded with descendants in an effort to reconstruct the complete life history of each graduate.

There are obvious limitations in relying on Yale graduates to infer the

mortality experience of New England's inhabitants. Infant and childhood

mortality cannot be measured. Perhaps the most serious limitation concerns the selectivity of the data. How representative of the general population were the Yale graduates? Although it is impossible to dismiss selectivity biases

completely, the possible distortions are far less significant than they first appear. Studies of preindustrial mortality in Europe consistently report that an elite advantage in life expectancy did not emerge until the late eighteenth


century, when rising inequality and urbanization eventually produced disease environments favoring upper classes (Johansson 1994; Woods and Williams

1995). Class differentials in mortality probably emerged much later in North America. The colonies were far less stratified by wealth than Europe, and their inhabitants benefited from wide access to land and food (Jones 1980; Williamson and Lindert 1980; McMahon 1985).3 Indeed, most studies find no significant mortality differentials by wealth or class before the late nineteenth century (D. S. Smith 1983a; Steckel 1988; Davin 1993; Lee 1997).4 We should not be surprised, therefore, to find that Yale graduates did not enjoy an advantage in life expectancy over that of their lower-status neighbors. Table 1 compares the life expectancy of Yale graduates at age 20, 30, and

50 with other estimates of early New England life expectancy. With few ex-

ceptions, life expectancies for Yale graduates were slightly lower than those

reported by other studies. Only in comparison with men in Nantucket and

Salem did Yale graduates live longer lives. The mortality experience of men in Nantucket, however, is undoubtedly unique due to occupational hazards associated with deep-sea whaling (Logue 1991), and the Salem study suffers from small sample sizes and erratic results (Vinovskis 1972).

Despite lingering concerns about selectivity biases, there are several

advantages to investigating the mortality experience of the Yale graduates. One advantage is the quality and completeness of the demographic data. Of the 2,307 men graduating from Yale College between 1701 and 1805 who survived to age 30, age at death is available for 2,284 (99.0%). Previous studies of early American mortality, in contrast, suffer from incomplete data and methodological flaws related to their use, limiting confidence in the results. A second advantage in the study of Yale graduates is the potential inclusion of explanatory variables. In addition to reporting the graduates' birth, graduation, and death dates, Dexter routinely notes their birthplace, residence, and occupation, and the number, date, and duration of their mar-

riages. Town of residence is available for 2,255 of the graduates who survive to age 30 (97.7%), migration history for 2,273 (98.5%), and occupation for

2,130 (92.3%).5 These data are ideally suited for the study of intraregional mortality determinants and for estimating the possible biases induced by migration censoring in existing estimates of early American mortality.


Table 1 Expectation of life in New England at various ages, by sex

Males Females Location, investigator, and period eo2 e30 eSo N(20) eo2 e30 eso N(20)a

Yale graduates and their wives (Hacker)

born 1680-99 born 1700-1719 bor 1720-39 born 1740-59 born 1760-79 born 1780-99

Andover, MA (Greven) born 1640-69 born 1670-99 born 1700-1729 born 1730-59

Family histories, Northeast (Pope)

born 1760-99

Hingham, MA (Smith) married 1641-1700 married 1701-20 married 1721-40 married 1741-60 married 1761-80 married 1781-1800 married 1801-20

Ipswich, MA (Norton)b married before 1700 married 1701-50

Nantucket, MA (Logue)C born 1660-1719 born 1720-59 born 1760-99

New England families

(Kasakoff and Adams)d born 1680-99 born 1700-1719 born 1720-39 born 1740-59

40.6 35.9 40.0 34.4 38.8 35.9 40.1 35.7 41.6 37.0 39.5 34.3

44.3 40.8 44.8 38.7 37.8 33.4 41.6 36.3

43.3 36.1

- 35.4 - 37.7 - 33.0 - 38.8 - 38.8 - 33.7 - 36.9

18.0 21.1 22.7 22.8 24.0 21.6

23.5 23.5 24.2 24.5

(86) (339) (509) (624) (725) (251)

(92) (192) (198) (88)

37.8 39.5 38.3 40.5 40.8 42.8

42.1 42.1 39.5 43.1

34.2 34.7 33.8 35.0 36.3 36.6

33.8 35.9 36.5 36.5

19.8 25.1 22.9 22.8 23.5 24.1

21.3 22.4 25.8 25.0

(73) (285) (499) (665) (742) (438)

(44) (108) (136) (36)

- (276) 45.8 39.1 - (151)

20.4 25.1 22.7 24.8 24.5 23.0 21.9

45.0 - 23.1 39.9 32.3 19.8

(77) -

(56) -

(65) -

(97) -

(107) -

(128) -

(168) -

n.a. 46.3 n.a. 36.8

35.0 28.1 35.0 37.3 34.4 39.2 40.7

23.2 20.7 21.9 24.5 25.2 23.8 25.0

- 22.9 29.9 18.7

(77) (50) (58) (97)

(102) (109) (149)

n.a. n.a.

44.1 37.5 24.0 (230) 47.5 40.0 24.9 (200) 38.5 33.8 22.1 (570) 43.3 36.5 23.4 (560) 35.1 32.6 21.7 (770) 41.2 36.6 23.0 (750)

40.6 43.9 42.6 46.6

(41) (75)

(106) (211)


Table 1 Continued

Males Females Location, investigator, and period e20 e 30 e50 N(20) e20 e30 e 5 N(20) a

born 1760-79 45.3 - - (337) - - -

born 1780-99 44.0 - - (586) - - -

Plymouth colony (Demos)e seventeenth century 48.2 40.0 31.2 (323) 41.4 34.7 29.7 (322)

Salem, MA (Sommerville)f seventeenth century 36.1 29.2 19.1 n.a. 21.4 20.0 14.4 n.a.

eighteenth century 35.5 30.3 19.6 n.a. 37.0 32.6 21.1 n.a.

Sources: Data from Dexter 1885-1912; Vinovskis 1972; Pope 1992; Smith 1973; Logue 1991; Kasakoff and Adams 1995. aN(20) is the sample size at age 20; e20 is years of remaining life expected at age 20, and so on. be20 for males and females is life expectation at age 21.

CRough estimates of sample sizes at age 20 based on person-years at risk. dLife expectancy at age 20 estimated from mean age at death figures. eCombined sample size of 645 divided evenly between men and women. fe20 for males and females is life expectation at age 21, e30 at age 31, and so on.

Intraregional Determinants of Adult Mortality

Although Yale graduates generally shared similar, upper-class backgrounds, they lived fairly diverse lives. Most settled in rural Connecticut towns, but

enough lived in urban areas and other colonies to provide contrasting mor-

tality experiences. College graduates generally pursued professional careers, but occupations varied within this class. One in three Yale graduates became a minister, one in four a lawyer, and another one in four a doctor or a merchant. Most graduates surviving to old age eventually married, but 445 died having never married. Given their similar backgrounds, mortality differentials observed within the Yale graduate population could be assumed to arise from the graduates' postgraduation life experiences, not from initial differences.

Using event-history techniques, this section estimates the impact of residence type (urban/rural and coast/interior), residence colony, migration history, occupation, and marital status on graduate survival.6 Mean and median survivals from age 30 are compared for various groups, and differences in their survival functions are tested for statistical significance with the log-


rank statistic. In addition, proportional hazard models are constructed to control for potential covariance among the factors.7

Urban/Rural and Coast/Interior Residence

Crude death rates in early New England varied markedly from town to town.

High rates of natural population growth, dispersed settlement patterns, and a low level of background mortality resulted in infrequent but intense out- breaks of infectious diseases that struck some villages with tremendous force and missed others completely (Dobson 1989; Caulfield 1938; Duffy 1953). Given that smaller, isolated villages were more likely to escape epidemics, it is perhaps reasonable to assume that these short-term differentials in mor-

tality led to substantial advantages in the average duration of life. In his classic synthesis of early New England mortality studies, Maris Vinovskis ar-

gued that residents of rural towns, especially those isolated from major trade

routes, enjoyed greater life expectancy than inhabitants of more densely populated urban centers, although the differential became less significant in the nineteenth century (1972). Recent analysis of crude death rates by Mary Dobson seems to confirm his hypothesis (1989), but the existence of a significant urban/rural differential in life expectancy remains inconclusive due to possible differences in death registration, population estimates, and

age structure between urban and rural areas.8 Life expectancy estimates are limited to a few small, agricultural towns; large urban centers such as New York and Boston have not been studied. Even comparisons of life expectancy among the existing rural town studies are risky; early researchers relied on small samples and did not explicate their methods or describe their sources.

The Yale data provide a unique opportunity to measure the impact of residence type on adult survival. Most graduates, of course, lived in the small towns and rural areas that comprised the overwhelming bulk of the

early American landscape. But enough graduates lived in urban locations to allow us to compare their mortality experience with that of their rural classmates. We can also compare the mortality experience of graduates living on the coast or near port towns, whether urban or rural, with that of graduates living in the interior.9

Table 2 depicts the results of three Kaplan-Meier survival analyses of the Yale data. Mean and median survival from age 30 is given for each factor.


Table 2 Kaplan-Meier survival analyses, graduates of Yale College, 1701-1805, surviving to age 30, by residence type

Mean Median Residence at age 30 survival (years) survival (years) Significance Na

Urban/rural residence 0.17 Urban 32.5 35.0 454 Rural 34.3 36.3 1,853

Coast/interior residence 0.10 Coast 32.8 35.0 811 Interior 34.5 37.0 1,496

Urban/rural residence Large urbanb 32.8 35.5 0.18 96 Small urban 32.4 34.4 0.34 358 Rural 34.3 36.3 reference 1,853

Source: Data from Dexter 1885-1912. aN is the total number of graduates participating in each factor. Approximately 1% of the graduates exit observation before dying (right-censored). bNew York City, Philadelphia, and Boston.

Results of the analyses support the hypothesized relationships. Graduates

living in rural areas survived on average 1.8 years longer than their urban

classmates, and those living in the interior enjoyed a 1.7-year survival ad-

vantage over those living near the coast. The statistical test for difference between the survival functions, however, is insignificant, so we cannot reject the null hypothesis that graduate survival by residence type was the same.

Moreover, the absolute differentials in survival are not very large. Studies of

nineteenth-century mortality indicate much larger urban/rural differentials

(Jaffe and Lourie 1942; Vinovskis 1972).10 Several factors may explain the smaller-than-expected urban/rural dif-

ferentials. First, most graduates living in urban areas were living in small to medium-sized cities such as New Haven, Norwich, New London, and Hartford. Only 13 graduates were living in Boston at age 30. New York and

Philadelphia, the other large urban cities in the United States circa 1790, were home to just 67 and 16 graduates, respectively. Perhaps the smaller cities that were home to the majority of Yale graduates did not represent significant health risks. The third survival analysis shown in Table 2, how-

ever, reveals little difference in survival between graduates living in large


cities (Boston, New York, and Philadelphia) and small cities. Mean survival in small cities (32.4 years) was slightly below that of large cities (32.7), and

neither of their survival functions were found to be significantly different

from the survival function of graduates living in rural areas. Second, the

severity of the public health hazards associated with cities-poor sanitation, water, and crowded housing-may not have become critical until urbanization and immigration accelerated in the nineteenth century (Blake 1959;

Duffy 1990). In 1790 New York City had a population of just 32,305, and New Haven and Hartford had populations of less than 5,000. By 1860, when

significant urban/rural differentials are known to exist, New York's population had increased to 805,658, New Haven's to 39,267, and Hartford's to

29,152 (U.S. Census Office 1990 [1802]; U.S. Census Office 1990 [1864]). A third possibility for the smaller-than-expected urban/rural differen-

tial in survival is the analytic treatment of urban/rural status as a constant, rather than a time-varying factor. A small percentage of the graduates living in rural areas at age 30 moved to urban areas later in life, and a small

percentage of the graduates living in urban areas migrated to rural areas.

Although the numbers of graduates changing residence type are small, they tend to reduce the absolute differential observed." A fourth possibility is

that, although they did not enjoy an overall advantage in life expectancy, elite

populations were well adapted to survival in urban environments. Richard and Claudia Bushman, for instance, argue that modern notions of cleanliness were first adopted by wealthier segments of society beginning in the eighteenth century (1988). Perhaps these newly acquired habits of personal hygiene mitigated the health risks associated with urban areas. Finally, another

possibility for the lack of a significant differential in urban/rural survival is this study's necessary focus on adult mortality, which may obscure larger and more significant differentials in infant and childhood mortality. Indi- viduals surviving to age 30 in urban areas may have already been exposed to several acute infectious diseases, immunizing them from subsequent dan-

ger. Meindl and Swedlund have observed that cohorts of individuals born in early-nineteenth-century western Massachusetts who were "stressed" in

childhood (through exposure to an unusually high level of infectious dis-

ease) subsequently enjoyed higher levels of adult survivorship than that of

unstressed, control cohorts (1977). This effect may have actually increased in

importance toward the end of the colonial period; Kunitz argues that many


acute infectious diseases had settled into an endemic pattern before 1800,

especially in densely populated areas, increasing the likelihood that urban adults would have been previously exposed (1984).

Colony of Residence

To date, all community-based estimates of New England life expectancy are for residents of small Massachusetts towns. Family-based genealogical studies also rely heavily on Massachusetts data. No life expectancy estimates are available for communities in Maine, New Hampshire, Vermont, or Con- necticut. Main, however, has used probate records to make rough estimates of life expectancy in seventeenth- and early-eighteenth-century Connecti- cut. His estimates of male life expectancy, ranging from 32 to 38 years at

age 21, are significantly lower than estimates for Massachusetts (Main 1985:

15). Although the use of different sources vitiates direct comparison, Main's results suggest the potential existence of large differentials in life expectancy among the New England colonies.

The Yale data provide a unique opportunity to estimate intercolony mortality differentials. While almost 60% of the graduates were living in Connecticut at age 30, enough were living elsewhere for comparison. Over 300 graduates were living in Massachusetts, an almost equal number were

living in New York or Long Island, and another 300 graduates were living in other colonies. Analysis of adult survival by colony, depicted in Table 3,

suggests that modest differentials in adult life expectancy existed among colonies, although the survival functions of graduates living in a few colonies were found not to be statistically significantly different from that of graduates residing in the reference colony of Connecticut. Roughly speaking, mean survival of graduates declined with latitude of residence. The longest mean survival from age 30, 38.6 years, was observed for graduates living in mis- cellaneous New England colonies (Maine, New Hampshire, Vermont, and Rhode Island). In order, mean survival then falls from Massachusetts (36.9), Connecticut (34.2), New York (32.7), Pennsylvania, Delaware, and New Jer- sey (32.2), to a low of 19.7 years for graduates living in the Chesapeake and Lower South. The results suggest that colony of residence (and its probable relationship to climate, disease patterns, and population density) played a small role in determining intraregional variations in adult survival.


Table 3 Kaplan-Meier survival analysis, graduates of Yale College, 1701-1805, surviving to age 30, by colony (state) of residence

Mean Median Residence at age 30 survival (years) survival (years) Significance Na

Residence colony/state Connecticut 34.2 36.1 reference 1,357 Massachusetts 36.9 40.0 0.06 312 New York 32.7 35.0 0.10 297 Other New England 38.6 41.0 0.00 136 Middle colonies/statesb 32.7 34.5 0.36 64 Southern colonies/statesc 19.9 17.7 0.00 71 Other and unknown 27.7 23.0 0.34 70

Source: Data from Dexter 1885-1912. aN is the total number of graduates participating in each factor. Approximately 1% of the graduates exit observation before dying (right-censored). bIncludes Pennsylvania, Delaware, and New Jersey. CIncludes Maryland, Virginia, North Carolina, South Carolina, Georgia, Florida, Alabama, Mississippi, and Louisiana.

While the focus of this investigation is on intraregional determinants of New England mortality, the relatively low mean survival of graduates living in southern colonies generally agrees with the demographic literature on the Lower South and Chesapeake regions, which reports very low estimates of adult life expectancy (Walsh and Menard 1974; D. B. Smith 1978; Rutman and Rutman 1979; Gallman 1980; Levy 1987). Some of the differential in the Yale population, however, may be due to "seasoning" mortality. Of the 71 Yale graduates known to be living in southern colonies at age 30, 48 were born in either Connecticut, Massachusetts, or New York. There is also reason to be cautious about the existing estimates of southern life expectancy. Most estimates are derived from reconstitutions of seventeenth-century counties. Record quality is an acknowledged weakness, and the additional problem of moderate levels of geographic mobility (Rutman and Rutman 1980) results in a very low percentage of individuals being successfully reconstituted. Per-

haps the best mortality data on the southern colonies, Daniel Levy's research on Maryland and South Carolina legislators, suggests that adult life expectancy increased rapidly in the colonial period. Life expectancy of Maryland legislators at age 30 rose steadily, from 17.5 in the 1635-49 birth cohort to 28.1 in the 1750-64 cohort. Life expectancy at age 30 also rose for South


Table 4 Kaplan-Meier survival analysis, graduates of Yale College, 1701-1805, surviving to age 30, by migration history

Mean Median survival (years) survival (years) Significance Na

Migration status at age 30 Resident town of birth 34.0 36.0 reference 749

Intracolony/state migrant 34.2 36.4 0.92 826 Intercolony/state migrant 34.1 36.5 0.87 698 Unknown 21.1 18.0 0.00 34

Lifetime migration b Resident town of birth 57.4 60.0 reference 664

Intracolony/state migrant 61.7 65.0 0.09 735 Intercolony/state migrant 62.7 66.0 0.00 1,070 Unknown 44.1 38.0 0.00 84

Source: Data from Dexter 1885-1912.

aN is the total number of graduates participating in each factor. Approximately 1% of the graduates exit

observation before dying (right-censored). bSurvival from birth for all graduates, by lifetime migration.

Carolina legislators (although more erratically), from 24.9 in the 1650-99 birth cohort to 27.5 in the 1750-64 birth cohort (Levy 1987, 1996). The Yale graduate survival results also support the mortality literature on the middle colonies, which reports male life expectancies at age 30 in the low

30s, just slightly below most estimates for New England (Kantrow 1989; Henderson 1990).

Migration History

Because the Yale data document each graduate's lifetime migrations and resi-

dences, we can test whether nonmigrants enjoyed an advantage in survival over that of migrants. Survival analysis reveals that less than a quarter-year differential in mean survival separates graduates residing in the town of their birth at age 30 from those migrating between colonies or within their natal

colony (Table 4).12 Moreover, the statistical test for difference between the survival functions of "stayers" and "movers" proved insignificant, so we cannot reject the null hypothesis that survival of migrants and nonmigrants was equal. Only when lifetime migration is considered does migration seem


Table 5 Kaplan-Meier survival analysis, graduates of Yale College, 1701-1805, surviving to age 30, by occupation

Mean Median survival (years) survival (years) Significance Na

Age 30 occupation Minister 36.1 38.2 reference 795 Lawyer 33.9 36.0 0.01 491 Merchant 32.5 35.0 0.00 271 Doctor 32.3 34.0 0.07 245 Public servant 34.7 37.0 0.66 98 Farmer 38.8 38.0 0.63 90 Teacher 34.3 37.8 0.60 80 Other or unknown 27.6 27.0 0.00 237

Source: Data from Dexter 1885-1912. aN is the total number of graduates participating in each factor. Approximately 1% of the graduates exit observation before dying (right-censored).

to play a significant role in determining graduate survival, with migrants outliving nonmigrants by over five years. The difference, however, is almost

certainly due to problems of circularity and selection. Because the cumulative chance of having migrated increases with age, graduates living longer lives were more likely to have migrated than graduates dying young. We should not be surprised, therefore, to find that a graduate dying at age 30 was more likely to die in his town of birth than a graduate surviving to age 90. Differences in survival between lifetime movers and stayers implies that

community-based reconstitution studies (which rely exclusively on lifetime

stayers) may be subject to substantial biases, discussed in more detail later.

Occupation

Occupational differences in survival are compared in Table 5. The longest mean survival from age 30, an additional 38.8 years, was observed for

graduates pursuing farming and agricultural careers. Ministers survived 36.1

years on average, followed by public servants (34.7), teachers (34.2), lawyers (33.9), merchants (32.5), and doctors (32.3).13 The apparent differential between farmers and professionals should be interpreted with caution. Most

eighteenth-century graduates, even those pursuing a professional calling,


probably operated a farm. Only those graduates explicitly mentioned by Dexter to be operating a farm and with no other known profession were classified as farmers. Their relative rarity in the records increases the probability that their estimated advantage in survival was due to chance variations in the data.14 In fact, as the reported level of significance indicates, the survival function of farmers was found not to be significantly different from that of ministers.

Marital Status

Demographic studies of modern populations consistently show that married men live longer than unmarried men. Two categories of hypotheses attempt to explain the relative advantage enjoyed by those who marry. Marriage protection emphasizes the potential health benefits that individuals receive from

marriage as a result of various social, psychological, or economic factors. Several studies, for instance, observe that marriage discourages risky be-

haviors, and others report that married individuals benefit from the sharing of economic resources and a division of labor. The presence of a spouse also may aid in recovery from illness or poor health through the recep- tion of better care. Marriage selection presumes that healthier individuals are more likely to marry in the first place. Individuals with observably poor health or unhealthy behaviors such as alcohol or substance abuse may find it more difficult to attract a spouse. Potential spouses also may be excluded from marriage through a wide range of selection criteria, including poverty, emotional instability, and physical unattractiveness, some of which may be correlated with higher mortality.

Despite the robust relationship between marital status and mortality in the twentieth century, no study of early American mortality has inves-

tigated the mortality experience of never-married individuals. The neglect is probably attributable to the convention of using marriage to establish the

population "at risk." Because Yale graduates enter the at-risk population at

graduation, however, it is possible to evaluate the impact of marital status on adult mortality. Of the 2,553 men who graduated from Yale College between 1701 and 1805, only 12 were married before obtaining their degree, which occurred at the mean age of 21.0. Most graduates delayed marriage until they were well established in a professional career, resulting in a mean


Table 6 Life expectancies of the graduates of Yale College, 1701-1805, by marital status, with age-specific relative mortality ratios

Never-married graduatesa Ever-married graduates RMRb

At exact Person-years Person-years age x at risk at risk

(x,x + 5) ex (x,x +5) ex (x,x + 5)

20 7,366 32.6 1,138 42.2 1.12 25 6,085 29.2 5,272 39.2 2.25 30 2,396 26.1 8,720 35.3 2.27 35 1,314 23.5 9,179 31.7 2.49 40 876 21.6 8,937 28.3 2.24 45 661 19.2 8,553 24.9 2.54 50 533 17.0 7,910 21.3 2.15 55 397 15.2 7,149 18.1 2.84 60 293 15.2 6,226 15.0 0.86 65 251 12.0 5,100 12.2 0.96 70 194 9.4 3,879 9.7 0.93 75 123 6.9 2,626 7.6 1.28

Source: Data from Dexter 1885-1912. aEver-married graduates are treated as part of the never-married population until the age at which they first marry. bRMR= 5m xnever-married graduates/ 5m ever-married graduates

age at marriage of 27.8. By age 30, 72% of the surviving graduates could be classified as ever married, increasing to 90% at age 40 and 93% at age 50. Of the 445 graduates not known to have married, over 50% died before

reaching their 32nd birthday, suggesting the possibility that many of these men would have married had they survived longer.

The time-varying character of marital status presents a difficult meth-

odological problem for Kaplan-Meier survival analysis. Of the 648 Yale

graduates who had never married by age 30, 408 eventually did so, leaving only 236 who remained never married at the time of their death. By age 50, marriage and death had reduced the never-married population to just 118 graduates. Rather than attempting to control for this bias by assum-

ing that never-married graduates are right-censored at marriage, I simply constructed life tables for both ever-married and never-married graduates with precise control of the population at risk.15 The results depict sizable differences in life expectancy by marital status (Table 6). The differential,


approximately 10 years at ages 20-24 and 25-29, falls rapidly after ages 30- 34 but is still evident at ages 55-59, at which point ever-married graduates could still expect to live 3 years longer than their never-married classmates.

Table 6 also includes age-specific measures of the relative mortality ratio

(RMR)-the ratio of the death rate of never-married graduates to ever- married graduates. The age pattern of the RMR is similar to that reported for modern populations. The RMR climbs steeply from near unity at ages 20-24 to over two at ages 25-29, remains relatively level through ages 55-59, and then falls rapidly back to near one at older ages. A few demographers have

argued that this pattern is evidence of the relative importance of marriage selectivity factors. As the percentage of the population that remains never married declines, it includes an increasing proportion of persons with serious health problems. Thus, excess mortality of the single population is low at younger ages when it includes many healthy individuals, increases up to the age at which few additional single persons marry, then remains relatively constant until health problems not originally present at marriage develop in the ever-married population (Livi-Bacci 1985; Kisker and Goldman 1987). The Yale pattern is somewhat unique, however, in that the RMR comes close to its peak value before half of the population is married, and it remains

relatively constant well beyond age 40, after which very few first marriages take place. Given the relatively high mortality environment of the eighteenth century, one would expect RMR values to fall as the ever-married population developed serious health problems. Thus, an argument can be made that the age-pattern of RMR derived from the Yale graduates could not arise

solely from the effect of marriage selection and may be based in part on mar-

riage protection factors. A recent simulation by Goldman (1993), however, suggests that inferences about the relationship between marriage selection and the observed patterns of the relative mortality ratio may be theoretically unsound. Marriage selection mechanisms acting alone can easily produce counterintuitive age patterns of RMR, including ratios that do not decline at the end of marriage or even within the human life span.

Regardless of the relative importance of marriage protection and mar-

riage selection factors, the Yale data confirm that marital status was a significant determinant of male life expectancy in early America. Therefore, the exclusion of never-married individuals from most early American mortality studies probably biases reported life expectancies at younger ages upward. If


we assume that the Yale population was representative of the general population in terms of the proportions marrying and the marital status mortality differential, the exclusion of never-married individuals from these studies results in overestimating male life expectancy at age 20 by about two years.

Multivariate Hazard Analysis

In the preceding survival analyses, explanatory variables were tested sepa- rately. The possibility exists, however, that the variables covary or interact,

lessening the estimated effect of some variables and increasing the effect of others. Ministers, for instance, may live longer lives than graduates pursuing other professional careers because of health benefits associated with their

residence, not because of a lower mortality risk associated with their occu-

pation. Ministers tended to live in rural areas, whereas professionals were more likely to live in urban locations. On the other hand, early graduates of Yale College almost invariably pursued ministerial careers. Later classes were more diverse, including an increasing proportion of graduates who became doctors and lawyers. Long-term trends in life expectancy could mask what may have been a greater occupational advantage for ministers.

Proportional hazard models (Cox regression) allow us to control for

potential interactions among multiple independent variables, or covariates

(Allison 1984). Hazard models are analogous to multivariate regression models. Unlike typical regression models, however, hazard models can incorpo- rate cases with censored observations. Hazard models can also accommodate

time-varying covariates, allowing us to control for conditions that change over a graduate's lifetime. For categorical variables, the estimated parameter coefficients represent deviations from the hazard rate of the omitted cate-

gory. The larger the hazard coefficient on a parameter, the sooner the event

(death) occurs. Table 7 presents the results of two hazard models. Model 1 evaluates the

effects of age, period, and marital status on the mortality rate of Yale graduates who survived to age 30. Model 2 adds residence colony, urban/rural residence, migration status, and occupation. Unlike age, period, and marital status, which are treated as time-varying covariates, these variables are treated as simple static variables.16 All covariates in Models 1 and 2-in-

cluding age-are categorical variables and are coded as dummy variables;


Table 7 Hazard model of the effects of age, period, marital status, residence

colony/state, urban/rural residence, migrant status, and occupation on the death rate of graduates of Yale College, 1701-1805 (Cox partial likelihood model)

Independent variable

Model 1 Model 2

Estimated Relative Level Estimated Relative Level coeff. (B) risk (eB) of sig.a coeff.(B) risk (eB) of sig.a

Age 30-39 40-49 50-59 60-69 70-79 80+

Period 1740-59 1760-79 1780-99 1800-1819 1820-39

Marital status Never married Married Widowed

Age 30 residence

colony/state Connecticut Massachusetts Other New England New York All other

colonies/states Age 30 residence type

Rural Urban

Age 30 migrant status Resident natal town

Intracolony/state migrant

Intercolony/state migrant

-0.31

reference 0.53 1.11 1.84 2.55

reference -0.10 -0.29 -0.30 -0.15

0.58

reference 0.16

0.73 0.001 -0.32 1.00 reference 1.70 0.000 0.54 3.03 0.000 1.13 6.27 0.000 1.87

12.78 0.000 2.59

1.00 reference 0.90 0.380 -0.12 0.74 0.007 -0.31 0.74 0.005 -0.31 0.86 0.185 -0.18

1.78 0.000 1.00 1.18 0.024

0.53

reference 0.15

reference -0.11 -0.26

0.02

0.26 1.29 0.007

reference 1.00

0.01 1.01 0.921

reference 1.00

0.15 1.17 0.016

0.13 1.14 0.110

0.73 1.00 1.72 3.10 6.52

13.30

1.00 0.89 0.73 0.73 0.84

1.70 1.00 1.16

1.00 0.90 0.77 1.02

0.001

0.000 0.000 0.000 0.000

0.310 0.005 0.004 0.121

0.000

0.043

0.160 0.039 0.868


Table 7 Continued

Model 1 Model 2

Independent Estimated Relative Level Estimated Relative Level variable coeff. (B) risk (eB) of sig.a coeff.(B) risk (e8) of sig.a

Age 30 occupation Minister reference 1.00

Lawyer 0.17 1.19 0.025 Doctor 0.17 1.19 0.040 All other 0.21 1.24 0.002

Source: Data from Dexter 1885-1912. Note: This table shows the results of 11,678 intervals of observation from 2,204 graduates. Only graduates surviving to age 30 and living in the period 1740-1839 are included in the analysis. During this

period graduates contribute 66,808 years at risk and 1,728 deaths.

aProbability of obtaining the estimated coefficient under the null hypothesis that the true coefficient is zero.

the excluded value functions as a reference group. Both models support the known relationship of age and mortality risk. The estimated coefficient for

graduates age 30-39 is negative, indicating that the hazard rate of death was below that of graduates in the age 40-49 reference group, and the estimated

parameters for all graduates in older age groups are positive, indicating higher hazard rates. More intuitively, the "relative risk" column, obtained by exponentiating the estimated coefficient, indicates the risk of death relative to the reference category. In Model 1, the relative risk of death was approximately 27% lower for graduates age 30-39 than for graduates in the age 40-49 reference group, 70% higher for graduates age 50-59, 303% higher for graduates age 60-69, and increasingly higher for graduates in older age groups.

A significant period effect is evident in both models. The negative coefficients indicate that periods after the 1740-59 reference period witnessed a decline in mortality risk. The decline was not steady, however. The relative risk of death was approximately 10% lower in the period 1760-79 than in the reference period, 25% lower in 1780-99 and 1800-1819, but only 15% lower in the period 1820-39. Significance testing suggests that only the periods 1780-99 and 1800-1819 were significantly different from the mid-eighteenth- century reference period. These results suggest that Yale graduates enjoyed a decline in mortality after 1780 that lasted into the nineteenth century. After

1820, however, mortality increased to near its mid-eighteenth-century level.l7


A strong marital status effect is also evident in both models. Married

graduates enjoyed a much lower hazard rate of death than that of single or widowed graduates. Compared to married graduates, single graduates suffered a 78% greater relative risk of death and widowed graduates an 18%

greater risk. The larger differential for single graduates probably reflects

marriage selectivity factors. The relatively smaller difference for widowed

graduates probably reflects marriage protection factors (such as the absence of care) or results from mortality clustering within families, although mar-

riage selection factors may also affect the propensity of widowed graduates to remarry.

Model 2 confirms many of the observations made with univariate survival analysis. Graduates living in the reference colony of Connecticut ex-

perienced a somewhat higher hazard risk of death than graduates living in other New England colonies, an approximately equal risk to those living in New York, and a lower risk than those living in more southern colonies. Despite controlling for multiple independent variables, urban/rural residence remains an insignificant determinant of adult mortality. Nor does multivariate analysis challenge the univariate survival analysis results on

occupational determinants. Ministers retain their advantage relative to those in other occupations. Migrant status, however, proves to be more significant than predicted by univariate survival analysis. Both intracolony and inter-

colony migrants have positive coefficients, indicating a higher hazard rate of death relative to graduates who lived in their natal town. These calculated coefficients are associated with a 17% and 15% higher relative risk of

death, respectively. Only the estimated coefficient for intracolony migrants was significantly different from that of the reference group of nonmigrants, however.

Using event-history techniques and the well-documented population of Yale

graduates, this section provided estimates of suspected determinants of

eighteenth-century mortality. It found that while some factors were sta-

tistically significant determinants of graduate survival, their impact was

generally modest. Marital status and colony of residence were the principal exceptions. Never-married graduates experienced a significantly greater risk of death than ever-married graduates, probably as a result of a combination of marriage selection and marriage protection factors. Intercolonial differen-


tials in survival were less pronounced but still significant. Graduates residing in Connecticut suffered a two- to four-year disadvantage in survival at age 30 from graduates living in other New England colonies. Urban/rural residence proved to be a weak determinant of adult survival. Graduates living in urban centers and near the coast suffered slightly higher mortality than

graduates living in rural towns and the interior, but the differences were small and possibly due to chance variations in the data.

The modest mortality differentials among colonies and the lack of significant urban/rural and coast/interior differentials suggest little diversity in New England's mortality experience. Short-term differentials in mortality, of

course, might occasionally be high, but long-term differentials and trends-

especially within the concentrated setting of most New England mortality studies- were probably small. Given the distinctive differences in mortality trends estimated by different methods noted later in this essay, the relatively small differentials within the Yale population suggest that discrepancies in the mortality literature are more likely to be due to small sample sizes and

methodological biases.

Biases in Early New England Mortality Studies

Investigations of early New England mortality generally can be classified as one of two types: community-based reconstitution or family-based genealogical studies.18 Since a death registration system was not established in New England until the late nineteenth century, all studies of early New

England mortality rely on nontraditional sources. Reconstitution studies at-

tempt to recover the vital events and family relationships of every resident in a selected community using a wide variety of surviving records. Based on

techniques developed by the French demographer Louis Henry in the 1950s

(Fleury and Henry 1956), these studies require years of laborious work link-

ing town, church, and graveyard records to reconstruct families and as much of a community's demographic history as possible (Greven 1967). In contrast to these source-intensive efforts, family-based genealogical investigations rely on published genealogies, selected for their quality and completeness. Many histories of New England families-typically researched and published by a descendant in the late nineteenth century-are readily available.


e 44 - \ / \ Andover, Mass. 0 -X Hingham, Mass.

Sa42 -s- -\ \y A Ipswich, Mass.

c,) \~sX \/ N / 4--U-Nantucket, Mass. . 40-

-e- Salem, Mass.

a 38

36--

34

1650 1675 1700 1725 1750 1775 1800

Estimated birth cohort

Figure 1 Male life expectancy at age 20 by estimated birth cohort, community-based reconstitution studies Sources: Data from Vinovskis 1972; Logue 1991. Notes: Where necessary, marriage cohorts have been converted to approximate birth cohorts and adult life expectancy at age 20 (e20) has been estimated from life expectancy at other ages. All observations are centered on the midyear of a given interval.

These genealogies can be sampled and combined to produce a database of thousands of individuals. Unlike community-based studies, which rely on

lifelong residents of the community for their mortality data, family-based genealogical studies include data on migrants.

A simple comparison of trends produced by reconstitution and genealogical studies makes a compelling case that the different methods, sources, and populations result in varying degrees of bias in their estimates. Figure 1, which depicts trends in adult male life expectancy reported by community- based reconstitution studies, and Figure 2, which depicts trends reported by family-based genealogical studies and the graduates of Yale College (which also includes data on migrants), present remarkably different pictures of early American mortality change. Community-based measures generally document a decline in eighteenth-century life expectancy, while genealogical studies report an increase.19


50

48-

46

I 44-

o C Kasakoff-Adams Sample 42 - - - Fogel Sample

/ ~~^ / Jr\cs --Yale Graduates

40 -

E 2 38-

36-

34 . . .

1650 1675 1700 1725 1750 1775 1800


Figure 2 Male life expectancy at age 20 by estimated birth cohort, family-based genealogical studies and graduates of Yale College, 1701-1805 Sources: Data from Kasakoff and Adams 1995: 205; Fogel 1986: 511; Dexter 1885-1912.

Notes: Life expectancy estimates for the Kasakoff-Adams sample have been calculated from mean ages at

death for men surviving to age 20 and are reported in 20-year overlapping birth cohorts. Fogel's 25-year

overlapping period estimates have been converted to approximate birth cohorts, and life expectancy at age 20 (e20) has been estimated from life expectancy at age 10. Life expectancy estimates for Yale graduates are reported in 20-year overlapping birth cohorts. All observations are centered on the midyear of a given interval.

Community-Based Reconstitution Studies

While community-based reconstitution studies have been invaluable in de-

scribing the salient characteristics of early New England's population-

especially its distinctive features of early marriage, high fertility, and rapid

population growth (D. S. Smith 1972; Wells 1992)--recent criticism sug-

gests that they are ill-suited for the study of mortality. The most significant

problems are related to small sample sizes and selectivity biases.20 Only indi-

viduals who marry and die in their natal town can be included in mortality

analysis. Migration after marriage results in at-risk individuals leaving the

area of study before their death is observed. Widows who remarry are also

at risk to fall out of observation, as records are linked nominally. Given

the additional problem of death under-registration, which is estimated to

have averaged about 50% (Vinovskis 1972: 196), most studies manage to


determine the age at death for only a small fraction of the community's inhabitants. When life expectancy estimates are reported by sex and birth

cohort, the number of cases is often too small to express much confidence in the results. Greven's estimate of male mortality in the 1740-59 birth cohort, for example, is based on the age at death of just 88 men (Greven 1970). In-

deed, the small number of "reconstituted" individuals in most studies may alone explain discrepancies in the mortality literature (see Table 1).

Selectivity biases are perhaps an even greater concern. Researchers have

long questioned the representativeness of the reconstituted population in

community-based studies. In addition to their failure to migrate, reconstituted individuals are more likely to have kin in the community, own land, and be married than the general population -factors that may be associated with greater longevity. Using a computer simulation of English reconstitution studies, however, Steven Ruggles has recently demonstrated that the

systematic exclusion of migrants results in a severe downward bias in life

expectancy estimates, even if migrants and nonmigrants shared the same

age-specific mortality rates. Because the cumulative chance of migration increases with age, early deaths will always be over-represented among the known deaths. Many of the individuals who died in the community of their birth would have migrated had they lived longer. Ruggles concludes that

existing methods to counter the bias introduced by migration censoring, while better than no method at all, are inadequate for demographic regimes characterized by high mortality and moderate to high rates of migration (1992).21 The "Ruggles Effect" has been empirically demonstrated by Kasa- koff and Adams using New England genealogical sources. Mean age at death for men surviving to age 20 and dying in the town of their nativity was 8.2 years lower than that of men who moved at least once in their lifetime and 5.6 years lower than that of the population as a whole (Kasakoff and Adams 1995).

Although group differences in mortality between movers and stayers may act as a countervailing bias to migration censoring, Ruggles's study clearly suggests that trends in migration will spuriously affect trends in life expectancy reported by community-based studies.22 If out-migration rates were increasing in the eighteenth century, the reported decline in life

expectancy may be in error. In his reconstitution study of Andover, Mas- sachusetts, Philip Greven noted that out-migration was increasing in the


60

50 -

0 I*. \ A bQ -- --Yale Graduates

| Q r -A,c . \ - -Kasakoff & Adams 0E! \7^-jL ^^VSample 1

30 - - -- Linear Trend (Kasakoff& Adams sample)

'* \>-- -- -- Linear Trend (Yale 20 -- -graduates)

10

0 I Il t l

1650 1675 1700 1725 1750 1775 1800


Figure 3 Percentage of men residing in natal town at death by birth cohort Sources: Data from Kasakoff and Adams 1995: 205; Dexter 1885-1912.

eighteenth century (Greven 1970: 211-14). Migration data from the graduates of Yale College and Kasakoff and Adams's genealogical study suggest that the increase in out-migration was not unique to Andover. Both studies find evidence of a sharp increase in migration beginning with birth cohorts in the eighteenth century. Figure 3 plots the percentage of individuals in each sample by birth cohort who remained in their natal town until death. While Yale graduates were somewhat more likely to migrate than individuals in the genealogical sample, both groups experienced similar trends. Roughly speaking, over 40% of men born in the late seventeenth century remained their entire life in the town of their birth. By the end of the eighteenth cen-

tury the figure had fallen to approximately 25% (Kasakoff and Adams 1995; Adams and Kasakoff 1984).

The correlation between regional migration estimates and life expectancy trends reported by community-based studies is suggestive. As the

percentage of men dying outside their natal town increases, life expectancy estimates decline. Exactly how much of this decline is a result of migration censoring is, unfortunately, unknowable. Early research was not explicit about the method employed or the number and percentage of censored ob-


75

i 70-

-- Lifetime Stayers

jo 65 --- Lifetime Movers

g= I-"^-^ Combined

60

55

1700 1720 1740 1760 1780

Birth Cohort

Panel A Graduates of Yale College, 1701-1805

75

70 -

+-: ~~~/ ~~~ \ >^ "s -A Lifetime Stayers O 65 -- / -Lifetime Movers

- /Combined

60

55

1700 1720 1740 1760 1780

Birth Cohort

Panel B Kasakoff-Adams genealogical sample

Figure 4 Male mean age at death by birth cohort and lifetime migration status Sources: Data from Kasakoffand Adams 1995: 205; Dexter 1885-1912.

servations. A very rough guess can be hazarded, however. Figure 4 plots mean age at death by lifetime migration status and birth cohort for the Yale graduates (Panel A) and the men in Kasakoff and Adams's genealogical sample (Panel B). Both studies depict a widening gap between the mean age at death of migrants and nonmigrants as out-migration increased in the eighteenth century. Despite the fact that the overall mean age at death increased in both studies, age at death for the stayers alone declined. If the level of

migration and the migrant/nonmigrant age at death differential were simi-


lar among the reconstituted communities, the decline in eighteenth-century life expectancy reported by community-based studies may be in error. Life

expectancy estimates for later birth cohorts are biased downward anywhere from two to nine years. Given that most community-based reconstitution studies estimate less than a three-year decline in life expectancy over the course of the eighteenth century, these rough figures suggest that, rather than falling, life expectancy may have remained unchanged or even increased

moderately.

Family-Based Genealogical Studies

In theory, family-based genealogical studies of mortality are not subject to

migration censoring. The compiler presumably traced all family descendants

regardless of their residence location and number of moves. In practice, how-

ever, genealogical records also suffer from missing vital dates, some of which

may be due to migration censoring. In his recent investigation of nineteenth-

century mortality, Clayne Pope noted that 11% of the men in a typical family genealogy were missing birth dates, and 43% were missing death dates. The respective percentages for women were even higher: 16% were missing birth dates, and 64% were missing death dates. Moreover, the average rate of increase in genealogical populations is typically below the rate expected through natural increase, indicating substantial attrition of individuals from the study (Pope 1992). Regrettably, the two family-based genealogical studies that investigate eighteenth-century mortality exclude right-censored individuals (Fogel 1986; Kasakoff and Adams 1995). Therefore, while migration censoring is considerably less of a problem with genealogical samples than with community-based studies, censoring biases are still present.23

The criticism of genealogical records as a source for mortality studies is usually one of selection bias. The existence of genealogical information is a function of the number of descendants, which is itself related to high life

expectancy, low age at marriage, and wealth (D. S. Smith 1979; Willigan and

Lynch 1982: 111-12). Fogel, however, found close correspondence between

period life expectancy estimates of the genealogical population and estimates made by Haines with the 1850 and 1860 censuses of mortality (1986: 448-55). By linking their sample to the 1850 census, Adams and Kasakoff determined that while the genealogical population was somewhat wealthier


than the general population, the difference was not large (1984). Selection biases may still exist in the seventeenth and eighteenth centuries, however, when most genealogies begin with the immigration of European forebears to the British colonies. Families experiencing high mortality in the first few

generations would have been much less likely to leave descendants. The dramatic differential in life expectancy between the family-based genealogical studies and the Yale graduates (Figure 2) supports this hypothesis.

If, in fact, genealogies overestimate life expectancy, the level of bias may change over time. One would suspect relatively less bias with generations closer to the compiler, although trends in migration, nuptiality, or the geographic composition of the sample may also affect reported trends. Selection biases may explain the dramatic 4- to 8-year increase in life expectancy re-

ported by the genealogical samples in the first 60 years of the eighteenth century. Given the already high level of early New England life expectancy, it is difficult to conceive of factors that could have caused so great a transfor- mation in mortality or of factors that could have improved the life chances of most New Englanders without also increasing those of the Yale graduates.

Clearly, more research is needed on the potential biases in genealogical records, especially in the eighteenth century. At the very least, attempts should be made to include right-censored individuals in life expectancy estimates and to utilize event-history techniques to estimate mortality determinants and possible biases.24 While genealogical records represent a great potential source for learning more about early American population patterns, the current evidence of a dramatic increase in eighteenth-century life

expectancy should be treated with caution. The evidence from community- based studies supports this hypothesis. Given even a worst-case scenario for biases induced by migration censoring, life expectancy estimates from

community-based studies suggest a much flatter trend than that estimated

by genealogical studies- a trend, in fact, closer to that observed for the

graduates of Yale College.

Conclusion

The study of mortality in early America lacks a clear consensus on mor-

tality trends and determinants. Roughly half the published studies report a

long-term decline in life expectancy; the other half document a long-term


increase or no secular trend. Using the rich data available for men graduating from Yale College between 1701 and 1805, this study tested two hypotheses that may explain the discrepancies. First, intraregional determinants of

mortality were estimated to determine the likelihood of a wide variety of

mortality levels and trends within New England. With the notable exceptions of marital status and regional residence, event-history analysis of suspected mortality covariates revealed a relatively "flat" mortality experience. No significant differential in survival was observed between graduates living in rural and urban areas or between those living in coast and interior communities. Small differentials in survival were apparent among graduates living in different New England colonies. But because most mortality studies rely on data drawn from Massachusetts, these small differences cannot explain the various levels and trends in mortality estimates. While one cannot completely discount the possibility that the different results reflect unique community experiences, the modest differentials observed within the Yale population suggest that long-term mortality patterns were similar across most of New

England.25 Instead, this study concludes that methodological biases explain much

of the variation in reported mortality trends. All estimates of early Ameri- can mortality suffer biases due to the fragmentary historical record. Be- cause most community-based estimates of life expectancy make no attempt to compensate for the effects of censoring biases, however, errors in these studies are probably the most severe. Since out-migration was increasing in the eighteenth century, life expectancy estimates from community-based studies probably suffer an increasing downward bias with later birth cohorts. A rough estimate of the potential biases suggests that once changing mi-

gration patterns are controlled for, life expectancy in most New England communities remained the same or increased moderately throughout the

eighteenth century. Family-based genealogical studies, though subject to far less migration censoring than community-based studies, are not immune to bias. While more research is needed on the potential biases in estimates derived from genealogical records, the dramatic increase in eighteenth-century life expectancy documented with these sources should be viewed with suspi- cion. The "corrected" trends in community-based studies and evidence from the graduates of Yale College-probably the most complete and clearly defined population studied in early America--suggest more moderate change.


This study concludes that there was no monotonic trend in early New En-

gland mortality. At the risk of implying more precision than is possible with the data, life expectancy probably increased moderately in the eighteenth century before declining slightly in the early part of the nineteenth century.

Notes

J. David Hacker is a graduate student in the Department of History at the University of Minnesota. His publications include "Cultural Demography: New England Deaths and the Puritan Perception of Risk," with Daniel Scott Smith, in the Journal of Inter-

disciplinary History (1996), and "Order out of Chaos: General Design of the Integrated Public Use Microdata Series," with Steven Ruggles and Matthew Sobek, in Historical Methods (1995). He wishes to thank Daniel Scott Smith, Steven Ruggles, Russell Menard, Matthew Sobek, Robert McCaa, members of the Early American Workshop at the Uni-

versity of Minnesota, and the anonymous reviewers and editors at Social Science History for comments on an earlier draft. This article was first presented at the annual meeting of the Social Science History Association in New Orleans, Louisiana, on 8 October 1996. 1 Although a consensus emerged that mortality increased in the eighteenth cen-

tury, the actual results were somewhat mixed. Using Greven's age at death data to construct abridged life tables, Vinovskis showed that generational declines in the average age at death for men and women in Andover did not translate into

steady declines in life expectancy by birth cohort, although the overall averages were lower for eighteenth-century cohorts than that for seventeenth-century cohorts. Smith's results for Hingham, Massachusetts, suggest that life expectancy was variable without trend in the eighteenth century. In a recent influential monograph on the social development of early America, however, Jack Greene concluded that "in New England, the deterioration of health conditions raised mortality to levels

by the mid-eighteenth century not too much lower than those in the contemporary Chesapeake" (Greene 1988: 171).

2 Because of the greater uncertainty associated with female life expectancy estimates, this essay focuses on reported trends and determinants of male life expectancy. Un- less otherwise noted, all figures and tables refer only to the male population. For an extended discussion of the importance of sex as a determinant of adult mortality and a critique of existing estimates see Hacker 1996: 63-116.

3 The relative advantage of British North Americans in access to land and food is

supported by anthropometric studies, which indicate that Americans were much taller than their contemporaries in England. Moreover, there is little indication that

height varied significantly by occupation, indicating that net nutrition was high for most early Americans (Sokoloff and Villaflor 1982; Sokoloff 1995; Steckel 1995).

4 Murray's recent study of late-nineteenth-century Amherst College graduates, which


reports higher-than-expected life expectancies, suggests that class differentials in

mortality emerged in the late nineteenth century (1997). 5 These data are described in more detail in Hacker 1996. 6 For an introduction to event-history techniques and examples of their application

to family reconstitution data see Gutmann and Alter 1993. A major benefit of

event-history analysis is the potential inclusion of "right-censored" cases. It is im-

portant to distinguish "right-censored" from "migration-censored" observations. While both terms refer to cases in which the event in question (death) is not ob-

served, event-history analysis assumes that censoring is noninformative about the event's occurrence. In other words, right-censored cases are assumed not to differ from those that remain under observation. An individual whose date of death is

unknown, for instance, must be assumed to experience the same age-specific mor-

tality rates after exiting observation as individuals whose dates of death are known.

"Migration censoring," in contrast, may or may not be informative. In fact, the self- selection of migrants and potential differences in the disease environment between

sending and receiving areas suggests that migrants experience different mortality rates from individuals who remain their entire lives in their natal town.

7 In addition to eliminating circularity, age 30 was chosen for a number of reasons.

Although mean age of graduation was in the early 20s (21.0), individual ages were

widely distributed. Thus, analysis of graduate survival from graduation, while repre- senting the original criterion for establishing risk and encompassing more years, is

complicated by the need to control for age at graduation. In addition, a few explanatory variables, such as residence and occupation, change rapidly for a few years after

graduation, after which time they become relatively stable. Graduates pursuing a ministerial career, for instance, tended to preach to several congregations before

being permanently "called," usually within a few years after graduation. Graduates

beginning medical careers either apprenticed for a few years with an established

physician or, in the later years of the study, enrolled in special medical colleges before

establishing themselves in practice at a needy community. By age 30 most graduates had established themselves in a lifelong occupation at a permanent residence.

All survival analyses were conducted with SPSS using Kaplan-Meier survival or Cox regression techniques. Mean and median survival are routinely estimated by SPSS with Kaplan-Meier analysis, which provides a display of mortality experience through the nonparametric estimates of the survival curve. Because distributions of survival times are typically skewed, median survival is perhaps the better measure of central tendency. I concentrate my discussion on mean survival, however, because of its closer relationship with life expectancy.

8 It is difficult to place much confidence in crude death rates for a number of reasons.

First, the completeness of colonial death registration is known to have varied over time and place, reducing confidence in the accuracy of the numerator (Gutman 1958). Rural towns are believed to have been more lax in registering deaths than urban areas. Susan Norton estimated that 59% of deaths were not registered in


colonial Ipswich, and Kenneth Lockridge estimated an under-registration of 44% in colonial Dedham (Vinovskis 1972: 204-5, 196). Second, because only a few scat- tered censuses were conducted in the colonial period, only rough estimates can be made about town populations, reducing confidence in the accuracy of the denominator (Wells 1975). Third, probable differences in age structure between urban and rural areas and between long-established and newly settled communities vitiate

comparisons. Eighteenth-century Boston, for instance, was said to be home to a

disproportionate number of widows. Interior settlements, in contrast, were characterized by a younger age structure (see quote by Cotton Mather in Cassedy 1969:

108-9). Even if life expectancies are equal, an older age structure will result in a

higher crude death rate. Finally, urban death rates are biased upward due to their inclusion of "seasoning" mortality among recent immigrants. So although a rural

advantage in life expectancy is believed to exist, the evidence is far from conclusive. 9 Instead of stepping into the methodological and conceptual quagmire of urban/rural

definitions, I simply relied on the urban/rural classifications reported in Cappon 1976: 97. An urban area is defined as a concentrated settlement of over 3,000 indi-

viduals, constituting an economic unit. In 1790 there were 21 places designated as

urban, including 4 cities in Connecticut. Each town in which graduates were living was designated urban or rural based on its status in 1790, roughly the halfway period of this study. While some towns designated urban in 1790 may have been defined as rural in 1740, and some designated rural would be urban in 1840, the variable

probably captures a greater degree of urbanization in these towns in all periods. I considered all graduates residing within 10 miles of the Atlantic seaboard as living on the coast, and all other graduates as living in the interior.

10 Jaffe and Lourie report large urban/rural differentials in 1825-35, but their work has been convincingly criticized by Vinovskis for relying on data from rural towns with implausibly low death rates. Using the relatively more complete death records available after midcentury, Vinovskis reports that males living in the 16 largest towns in Massachusetts in 1860 suffered about a five-year disadvantage in life expectancy at age 30. Rapid urbanization in the nineteenth century resulted in these towns'

having populations over 10,000, however. Less than a one-year differential in life ex-

pectancy separates towns with populations under 2,500 from those with populations ranging from 2,500 to 10,000 (Vinovskis 1972: 203-5, 211).

11 A survey of the 112 men graduating between 1745 and 1749 and surviving to age 30 revealed that a shift in residence status later in life was uncommon. Of the 88

graduates living in rural areas, only 2 moved to an urban location sometime later in

life, and just 4 of 24 graduates living in urban areas at age 30 subsequently moved to a rural location.

12 Graduates were classified as nonmigrants if they were living in their natal town at

age 30. Unless the graduate was born in New Haven, however, all graduates were

migrants in the sense that they lived for a few years in New Haven while attending college.


13 Graduates were classified with one occupation even if they pursued multiple careers.

Minister-physicians, more common in the early eighteenth century, were classified

simply as ministers, and graduates pursuing several occupations were classified, in order, as either a minister, lawyer, doctor, merchant, teacher, public servant, or farmer. Thus, a graduate employed in law and operating a farm or business was considered a lawyer. Most Yale graduates served a few years in some public capacity but were designated public servants only if no other evidence of a professional career was mentioned.

14 In their study of early New England families, Kasakoff and Adams report that men

engaged in farming died at older ages than men with nonfarm occupations, suggesting that the advantage in survival of farmers relative to other graduates with

professional occupations was not due to chance (Kasakoff and Adams 1995). 15 All graduates known to have ever married were treated as part of the never-married

population from the age they graduated until the age they married, at which point they were assumed to exit observation. They therefore contribute risk years, the denominator of the age-specific mortality rates, from graduation until marriage but do not contribute deaths to the numerator. Never-married graduates contribute risk

years from graduation to death and also contribute deaths to the numerator. In the calculation of life expectancy for ever-married graduates, individuals were not des-

ignated at risk until the age they married. Single graduates never contribute risk

years or deaths to the age-specific mortality calculations for married graduates. 16 Residence colony, urban/rural residence, migration status, and occupation were de-

termined for each graduate at age 30. A small number of graduates later moved to another colony or state, and a few changed residence type, but the percentage is believed to be small (see note 11). Unfortunately, time and funding constraints

precluded the entering of all residences in which the graduates lived; thus residence could not be treated as a time-varying covariate.

17 In general, period effects should be more marked than those for birth cohorts, although Kaplan-Meier survival analysis (not shown) indicates significant differentials in graduate survival at age 30 by birth cohort. The differential in survival between the 1680-1725 birth cohort and the 1725-74 reference birth cohort was

significant at the 0.01 level, and the 1774-89 birth cohort was significant at the 0.1 level.

18 There are a few exceptions. Most community-based reconstitution studies, such as Barbara Logue's investigation of Nantucket, rely on available genealogies to supplement local church and town records (1991). The investigation of Yale graduates is best described as a study of a special population-an approach popularized by Henry's investigation of the Genevan bourgeoisie and Hollingsworth's research on the British peerage (Henry 1956; Hollingsworth 1964). Daniel Levy's study of

Maryland legislators is an example of a special population in early America (1987). 19 Trends in adult female life expectancy reported by the community-based recon-

stitution studies vary so widely as to defy generalization. Female life expectancy


at age 20 is reported to have dropped precipitously by almost 10 years in Ipswich, skyrocketed more than 15 years in Salem, remained relatively stable in Andover, fallen by more than 8 years in Nantucket, and erratically risen (at age 25) more than 10 years in Hingham (see Table 1). The wide variety of results is probably the result of even smaller sample sizes, a greater percentage of missing data, and additional sources of bias. Women, for example, are likely to exit observation if they remarry, and they appear less often in wills and other colonial records, making it difficult to define the at-risk population precisely (Hacker 1996: 63-116).

20 There are additional problems with local studies. One weakness is the inherent de-

gree of speculation involved. The researcher must assume that similar names in different records-sometimes separated by several intervening decades-represent the same person. Children were often named after relatives with the same surname or deceased siblings, many of whose deaths may have escaped mention in surviving records, compounding errors of identification (Razzell 1993). A second problem is uneven geographic coverage. Most studies have concentrated on small, agricultural villages settled in the early seventeenth century. Communities settled at later dates remain relatively underexplored, and urban populations are simply too difficult to reconstitute (Vinovskis 1978). Third, the completeness of death registration declines over the course of the eighteenth century (Gutman 1958), resulting in possible cohort and period biases in mortality estimates. Finally, much of the early research has been criticized for methodological faults, and some of the published results seem inherently unbelievable. Demos's study (1965) is almost certainly biased due to its reliance on tombstone records, which have been shown to over-represent younger women and older men (Vinovskis 1978; D. S. Smith 1979; Gallman 1980; Hollingsworth 1969: 272-74).

21 Surprisingly, most New England mortality studies make no attempt to compensate for missing data or migration censoring. As a result, biases in reported life expectancies may be severe. Susan Norton's study of Ipswich, Massachusetts (1971), and Barbara Logue's study of Nantucket, Massachusetts (1991), are exceptions. Norton's

"high" and "low" estimates of adult survival, however, reveal the tenuous status of the results. Although the average survival estimates suggest that life expectancy was declining in Ipswich, the range between the high and low estimates is so wide that a number of other scenarios is possible, including an increase in life expectancy. Despite having to contend with lower-quality records--or perhaps for that very reason-Chesapeake mortality studies are much more sophisticated, routinely reporting high, low, and preferred mortality estimates. For the pioneering study, see Walsh and Menard 1974. Estimates of early New England migration conclude that migration after marriage was common in the eighteenth century (Norton 1973; Villaflor and Sokoloff 1982; Adams and Kasakoff 1984). Whether these rates were

"high" relative to other periods of American history, however, is uncertain (D. S. Smith 1983b).


22 Wrigley, while agreeing that migration censoring theoretically biases age at marriage downward, notes that empirical results in English reconstitution studies detect no

apparent effect, suggesting the existence of countervailing biases (Wrigley 1994). Kasakoffand Adams also suggest that group differences between migrants and non-

migrants account for some of the differential observed in the average age at death in their study of New England genealogies (Kasakoff and Adams 1995).

23 There are other potential biases in their life expectancy estimates. Although Fogel's study primarily relies on New England genealogies, families in other regions of British North America are also included. Moreover, the relative proportion of New

England families declines over the course of the study. Thus, trends in life expectancy reported by Fogel may reflect the changing geographic composition of the

sample. It should also be noted that life expectancy is only reported by period. While

having the advantage of pinpointing periods of mortality change, period estimates of life expectancy are not directly comparable to estimates by birth cohort. Kasakoff and Adams report only the average age at death for men in their sample who are known to achieve a given age, not life expectancy. While this measure can be used to calculate a useful proxy of life expectancy, life table techniques with precise control of the population at risk may produce different estimates for the various cohorts.

24 Kasakoff and Adams are currently in the process of adapting their sample for

event-history analysis (1996). 25 The experience of Nantucket, an island largely dedicated to commercial whaling,

is perhaps unique. The occupational hazards of deep-sea fishing were undoubtedly high and may have changed over time (Logue 1991).

References

Adams, J., and A. Kasakoff (1984) "Migration and the family in colonial New England: The view from genealogies." Journal of Family History 9: 24-42.

Allison, P. D. (1984) Event History Analysis: Regression for Longitudinal Event Data.

Beverly Hills, CA: Sage. Archer, R. (1990) "New England mosaic: A demographic analysis for the seventeenth

century." William and Mary Quarterly 47: 477-502.

Blake, J. B. (1959) Public Health in the Town of Boston, 1630-1822. Cambridge, MA: Harvard University Press.

Bushman, R. L., and C. Bushman (1988) "The early history of cleanliness in America."

Journal of American History 74: 1213-36.

Cappon, L. J., ed. (1976) Atlas of Early American History: The Revolutionary Era, 1760-1790. Princeton, NJ: Princeton University Press.

Cassedy, J. H. (1969) Demography in Early America: Beginnings of the Statistical Mind, 1600-1800. Cambridge, MA: Harvard University Press.


Caulfield, E. (1938) "A history of the terrible epidemic, vulgarly called the throat distem-

per, as it occurred in his majesty's New England colonies between 1735 and 1740." Yale Journal of Biology and Medicine 11: 219-72.

Davin, E. L. (1993) "The era of the common child: Egalitarian death in antebellum America." Mid-America 75: 134-63.

Demos, J. (1965) "Notes on life in Plymouth Colony." William and Mary Quarterly 25: 264-86.

Dexter, E B. (1885-1912) Biographical Sketches of the Graduates of Yale College, 1701- 1815. 6 vols. New York: H. Holt and Company.

Dobson, M. J. (1989) "Mortality gradients and disease exchanges: Comparisons from old

England and colonial America." Social History of Medicine 2: 259-97.

Duffy, J. (1953) Epidemics in Colonial America. Baton Rouge: Louisiana State Univer-

sity Press.

(1990) The Sanitarians: A History of American Public Health. Urbana: Univer-

sity of Illinois Press.

Fleury, M., and L. Henry (1956) Des registres paroissiaux a l'histoire de la population: Manuel de depouillement et d'exploitation de l'etat civil ancien. Paris: Editions d'Institut National d'Etudes Demographiques.

Fogel, R. W. (1983) "Secular changes in American and British stature and nutrition."

Journal of Interdisciplinary History 14: 445-81.

(1986) "Nutrition and the decline of mortality since 1700," in S. L. Engerman and R. E. Gallman (eds.) Long-Term Factors in American Economic Growth. Chicago: University of Chicago Press: 439-555.

Gallman, J. M. (1980) "Mortality among white males: Colonial North Carolina." Social Science History 4: 295-316.

Goldman, N. (1993) "Marriage selection and mortality patterns: Inferences and fallacies."

Demography 30: 189-208.

Greene, J. P. (1988) Pursuits of Happiness: The Social Development of Early Modern British Colonies and the Formation of American Culture. Chapel Hill: University of North Carolina Press.

Greven, P. J., Jr. (1967) "Historical demography and colonial America." William and

Mary Quarterly 24: 438-54.

(1970) Four Generations: Population, Land, and Family in Colonial Andover, Massachusetts. Ithaca, NY: Cornell University Press.

Gutman, R. (1958) "Birth and death registration in Massachusetts: The colonial back-

ground, 1639-1800." Milbank Memorial Fund Quarterly 36: 58-74.

Gutmann, M. P., and G. Alter (1993) "Family reconstitution as event-history analysis," in D. Reher and R. Schofield (eds.) Old and New Methods in Historical Demography. New York: Oxford University Press: 159-77.

Hacker, J. D. (1996) "Adult mortality in early New England: Evidence from the graduates of Yale College, 1701-1805." M.A. thesis, University of Minnesota.

Henderson, R. C. (1990) "Demographic patterns and family structure in eighteenth-


century Lancaster County, Pennsylvania." Pennsylvania Magazine of History and

Biography 114: 349-83.

Henry, L. (1956) Anciennes families genevoises, etude demographique: 16e-20e siecles. Paris: Presses Universitaires de France.

Hollingsworth, T. H. (1964) "The demography of the British peerage." Supplement to

Population Studies 18, no. 2.

(1969) Historical Demography. Ithaca, NY: Cornell University Press.

Jaffe, A. J., and W. I. Lourie, Jr. (1942) "An abridged life table for the white population of the United States in 1830." Human Biology 14: 352-71.

Johansson, S. R. (1994) "Food for thought: Rhetoric and reality in modern mortality history." Historical Methods 27: 101-25.

Jones, A. H. (1980) Wealth of a Nation to Be. New York: Columbia University Press.

Kantrow, L. (1989) "Life expectancy of the gentry in eighteenth- and nineteenth-century Philadelphia." Proceedings of the American Philosophical Society 133: 312-27.

Kasakoff, A. B., and J. W. Adams (1995) "The effect of migration on ages at vital events: A critique of family reconstitution in historical demography." European Journal of

Population 11: 199-242.

(1996) "Event history analysis of the effect of moving on adult survival: A view from New England genealogies." Paper delivered at the annual meeting of the Social Science History Association in New Orleans, 8 October.

Kisker, E. E., and N. Goldman (1987) "Perils of single life and benefits of marriage." Social Biology 34: 135-52.

Kunitz, S. J. (1984) "Mortality change in America, 1620-1920." Human Biology 56: 559-82.

Lee, C. (1997) "Socioeconomic background, disease, and mortality among Union Army recruits: Implications for economic and demographic history." Explorations in Eco- nomic History 34: 27-55.

Levy, D. S. (1987) "The life expectancies of colonial Maryland legislators." Historical Methods 20: 17-27.

(1996) "Long-run geographic and temporal changes in mortality in the colonial South." Paper delivered at the annual meeting of the Social Science History Association in New Orleans, 8 October.

Livi-Bacci, M. (1985) "Selectivity of marriage and mortality: Notes for future research," in N. Keyfitz (ed.) Population and Biology. Liege, Belgium: Ordina Editions: 99-108.

Logue, B. J. (1991) "In pursuit of prosperity: Disease and death in a Massachusetts commercial port, 1660-1850." Journal of Social History 25: 309-43.

Main, J. T. (1985) Society and Economy in Colonial Connecticut. Princeton, NJ: Prince- ton University Press.

McMahon, S. F (1985) "A comfortable subsistence: The changing composition of diet in rural New England, 1620-1840." William and Mary Quarterly 42: 26-65.

Meindl, R. S., and A. C. Swedlund (1977) "Secular trends in mortality in the Connecticut

Valley, 1700-1850." Human Biology 49: 389-414.


Murray, J. E. (1997) "Standards of the present for people of the past: Height, weight, and mortality among men of Amherst College, 1834-1949." Journal of Economic

History 57: 585-606.

Norton, S. (1971) "Population growth in colonial America: A study of Ipswich, Massa- chusetts." Population Studies 25: 433-52.

(1973) "Marital migration in Essex County, Massachusetts, in the colonial and

early Federal periods." Journal of Marriage and the Family 35: 406-18.

Pope, C. L. (1992) "Adult mortality in America before 1900: A view from family histo-

ries," in C. Goldin and H. Rockoff (eds.) Strategic Factors in Nineteenth-Century American Economic History: A Volume to Honor Robert W. Fogel. Chicago: Uni-

versity of Chicago Press: 267-96.

Razzell, P. (1993) "The growth of population in eighteenth-century England: A critical

reappraisal." Journal of Economic History 53: 743-71.

Ruggles, S. (1992) "Migration, marriage, and mortality: Correcting sources of bias in

English family reconstitutions." Population Studies 46: 507-22.

Rutman, D. B., and A. Rutman (1979) "Parental death in a seventeenth-century Vir-

ginia county," in T. W. Tate and D. L. Ammerman (eds.) The Chesapeake in the Seventeenth Century: Essays on Anglo-American Society. Chapel Hill: University of North Carolina Press: 153-82.

(1980) "'More True and Perfect Lists': The reconstruction of censuses for Middlesex County, Virginia, 1668-1704." Virginia Magazine of History and Biogra- phy 88: 37-77.

Smith, D. B. (1978) "Mortality and family in the colonial Chesapeake." Journal of

Interdisciplinary History 13: 735-59.

Smith, D. S. (1972) "The demographic history of Colonial New England." Journal of Economic History 32: 165-83.

(1973) "Population, family, and society in Hingham, Massachusetts." Ph.D. diss., University of California, Berkeley.

(1979) "The estimates of early American historical demographers: Two steps forward, one step back, what steps in the future?" Historical Methods 12: 24-38.

(1983a) "Differential mortality in the United States before 1900." Journal of

Interdisciplinary History 13: 735-59.

(1983b) "Migration of American colonial militiamen: A comparative note." Social Science History 7: 475-80.

Sokoloff, K. L. (1995) "The heights of Americans in three centuries: Some economic and

demographic implications," in J. Komlos (ed.) The Biological Standard of Living on Three Continents. Boulder, CO: Westview Press: 133-50.

Sokoloff, K. L., and G. C. Villaflor (1982) "The early achievement of modern stature in America." Social Science History 6: 453-81.

Steckel, R. H. (1988) "The health and mortality of women and children, 1850-1860."

Journal of Economic History 48: 333-45.


(1995) "Stature and the standard of living." Journal of Economic Literature 33: 1903-40.

U.S. Census Office (1990 [1802]) Return of the Whole Number of Persons within the Sev- eral Districts of the United States, According to "An Act Providing for the Second Census or Enumeration of the Inhabitants of the United States," Passed February Twenty-Eighth, One Thousand Eight Hundred. New York: Norman Ross.

(1990 [1864]) Population of the United States in 1860: Compiled from the

Original Returns of the Eighth Census. New York: Norman Ross.

Villaflor, G. C., and K. L. Sokoloff(1982) "Migration in colonial America: Evidence from the militia muster rolls." Social Science History 6: 539-70.

Vinovskis, M. A. (1972) "Mortality rates and trends in Massachusetts before 1860." Journal of Economic History 32: 184-218.

(1978) "Recent trends in American historical demography: Some methodological and conceptual considerations." Annual Reviews in Sociology 4: 603-27.

Walsh, L. S., and R. R. Menard (1974) "Death in the Chesapeake: Two life tables for men in early colonial Maryland." Maryland Historical Magazine 69: 211-27.

Wells, Robert V. (1975) The Population of the British Colonies in America before 1776: A Survey of Census Data. Princeton, NJ: Princeton University Press.

(1992) "The population of England's colonies in America: Old English or new Americans?" Population Studies 46: 85-102.

Williamson, J. G., and P. H. Lindert (1980) American Inequality: A Macroeconomic

History. New York: Academic Press.

Willigan, J. D., and K. A. Lynch (1982) Sources and Methods of Historical Demography. New York: Academic Press.

Wrigley, E. A. (1994) "The effect of migration on the estimation of marriage age in family reconstitution studies." Population Studies 48: 81-97.

Woods, R., and N. Williams (1995) "Must the gap widen before it can be narrowed?:

Long-term trends in social class mortality differentials." Continuity and Change 10: 105-37.

trends and determinants of adult mortality in early … and...trends and determinants of adult...

Documents