Recherches récentes Mes recherches
Vous consultezThe Family and Mortality: A Case Study from Rural Belgium
AuteursGeorge Alter du même auteurHistory and PIRT Indiana University Bloomingtom USA
A central problem in historical research on mortality decline is our inability to observe types of individual and family behavior that might affect mortality. Even the most successful studies of structural factors can explain only a small part of the variation in mortality (Reher and Schofield, 1991, 2; Pozzi and Robles Gonzales, 1996; Ryan Johansson and Kasakoff, 2000, 56). While economic conditions and public health measures can be identified with more or less precision, it is almost impossible to trace simple but essential actions that might have important consequences, like washing hands. Discussions of health practices in (elite) medical and public health circles can rarely be linked to the daily activities of common people (See Havelange, 1990, for an analysis of medicalization in the area under study here.) Nevertheless, we know that differences in individual behavior can have a strong influence on mortality. Demographic surveys consistently find that mother’s education is the most powerful predictor of infant and child mortality (Hobcraft, et al., 1984, 220; Caldwell, 1986, 184-7). It is by no means clear how better educated mothers actually protect their children, but even poor mothers can be successful (Preston and Haines, 1991). Studies of both contemporary and historical populations find that deaths tended to cluster in families for reasons that cannot be explained.
2 This study takes this last observation as its starting point. Differences in mortality among families give us clues about the importance of unobserved health-related behaviors. For example, if lower mortality was due to types of personal behavior learned in childhood, it should carry over to mortality at older ages. Thus, if lower infant mortality was due to better hygiene, we should also see lower mortality among children who were born in homes with fewer infant deaths. If people learned good health habits as children, then those who grew up in homes with lower infant and child mortality should have had lower mortality in adulthood. Of course, other factors affecting mortality may also persist from younger to older ages, such as wealth and environmental conditions, but we can control for these factors to some extent. The persistence of inter-family differences at older ages may give us a clue to the origins of mortality decline.
3 In this paper we use records from a nineteenth-century Belgian community to look at mortality differences among families in two ways. First, we construct a direct measure of exposure to disease in childhood by counting the number of children in each family that died before age 15. Second, we calculate the overall effect of inter-family differences by using a “random effect” model that estimates the variance of the “family effect”. Both of these measures show a strong family effect in childhood, but this effect diminishes after age 15 and disappears after age 55. Moreover, in a period still dominated by infectious diseases, those who survived diseases in childhood acquired immunities that helped them in later life. In particular, smallpox inoculation reduced the overall death rate, but it created a pool of susceptible adults during this transitional period.
4 The existence of inter-family differences in infant and child mortality is widely recognized. In both contemporary and historical studies deaths tend to cluster in families. Lynch and Greenhouse (1994) have shown the clear importance of this pattern in nineteenth-century Swedish records. Mortality studies conducted in the five societies examined by the Eur-Asia Project have also encountered this pattern (Bengtsson and Saito, 2000). In general, the best predictor of whether a child will survive is the survival of the preceding child. We have even seen a similar link between the mortality of mothers and children. If an infant died, its mother was at a greater risk of following it to the grave (Alter, Manfredini, and Nystedt, in preparation).
5 A number of factors contribute to this pattern. During childhood, siblings share the same living conditions and economic and social resources. However, there are strong indications that differences in parental behavior are important, at least from the beginning of the twentieth century. For example, it is well known that infant mortality was particularly low in Jewish families. Studies of the U.S. in the early twentieth century have found infant mortality in families of Jewish immigrants as low or lower than the mortality of native-born Americans, even though the incomes of the former were much lower (Woodbury, 1926; Condran and Kramarow, 1991; Hardy 1993; Ryan Johansson 2000, 62). Observing unusually low mortality among poor Jews in the nineteenth-century Venetian ghetto, Derosas (forthcoming) stresses the contrast between their cultural (and religious) commitment to health and life and the various forms of resignation in the Catholic population. (See also Knodel, 1988, for a similar discussion of infant mortality in Protestant and Catholic regions of eighteenth- and nineteenth-century Germany). What matters is not only knowledge but attitudes about disease prevention and interaction with the health system, an “instrumental rationality” (Thornton and Olson 1992) or “know-how” (Woods and Williams 1995, 130). Similarly, Das Gupta emphasizes the importance of “parental competence”: “‘Incompetent’ parents give their children poorer care, are slower to recognize and respond effectively to their needs, and consequently lose children.” (Das Gupta, 1990, 505).
6 Simple actions, like washing hands and utensils and avoiding polluted water or contaminated food, may have contributed to inter-family mortality differentials. The scientific case for better hygiene became more and more obvious after the experiments of Pasteur and Koch, but the importance of cleanliness was widely discussed even before acceptance of the germ theory of disease. James Riley (2001) has argued that “filth theory”, which linked disease to foul and decaying matter, had most of the same practical implications as germ theory. (See also Bernard Lecuyer (1986) and Alain Bideau and his colleagues (1988))
7 However, it is not clear how widely these ideas spread through various societies. Samuel Preston (1985) points to evidence that differences in infant mortality between well-educated (but not necessarily well-paid) professionals and less educated occupational groups expanded rapidly at the beginning of the twentieth century. He attributes this pattern to changing behavior among those who were the first to learn and accept the lessons of the germ theory.
8 Less is known about interfamily differences in adult mortality. Although infants and children are particularly sensitive to poor hygiene, these same behaviors may affect adult mortality as well. In this paper we pose a simple hypothesis, if health-promoting behaviors were important in adult mortality, then families with lower childhood mortality should have lower mortality after age 15 as well. The test of this hypothesis is based on two assumptions. First, we assume that health-related behaviors, such as personal hygiene, were learned during childhood and continued at later ages. Second, we assume that behaviors with a beneficial effect on the health of children should also benefit adults. Neither of these assumptions appears particularly controversial to us. The implication of this hypothesis is that a “family effect” on mortality should persist into adulthood and old age.
9 The persistence of health-promoting behaviors should result in a positive correlation between childhood mortality among siblings and mortality in adulthood, but there are at least two mechanisms that would result in the opposite relationship. First, exposure to some diseases produces immunity in later life, and diseases encountered in childhood are sometimes less fatal than the same disease experienced in later life. For instance, this is certainly the case for smallpox, a highly relevant example because our study area suffered a smallpox epidemic in 1871. Vaccination spread quite rapidly in Eastern Belgium at the beginning of the nineteenth century (Havelange, 1990, 264). How-ever, it was not immediately recognized that immunity from vaccination was temporary, while the immunity conferred by actually surviving an episode of smallpox is permanent (Pitkänen et al., 1989; Sköld, 1996, 479-80)
 Physicians in the Province of Liège did not begin...
10 A second alternative hypothesis is based on the assumption that individuals differ in their ability to survive disease (Vaupel, Manton, and Stallard, 1979; Vaupel and Yashin, 1985). At any level of mortality, those who are relatively more “frail” tend to die at younger ages, which makes the surviving population less “frail” as age increases. The composition of a population changes as it grows older, because the individuals most susceptible to disease tend to die first. When mortality is lower, the selective effect of mortality is reduced. At each age there are more “new survivors” who are relatively more frail. Thus, lower mortality at younger ages can have the apparently paradoxical effect of increasing mortality at older ages (Dupâquier, 1982). For our purposes, we would expect that individuals from low mortality families had experienced less selection than those in high mortality families. Consequently, the survivors of low mortality families would on average be more “frail” and more susceptible to disease at adult ages.
11 There is an ongoing debate in demography over the importance of selection effects due to heterogeneous “frailty”. It has been suggested that some unusual demographic patterns, such as mortality “cross-overs” can be explained by this process (Vaupel, Manton, and Stallard, 1979). For example, African-Americans have higher mortality than the rest of the U.S. population at younger ages but their old-age mortality is unusually low. Some analysts see this as a consequence of heterogeneous frailty, while others attribute it to age misreporting (Elo and Preston, 1997). While the hetero-geneous frailty hypothesis is based on a persuasive mathematical model (Riley and Alter, 1990; Mosley and Becker, 1991), it assumes the existence of an individual attribute that is impossible to observe or measure. It has, therefore, been very difficult to prove or disprove with empirical evidence.
12 To test the principle hypothesis posed here, it will be necessary to follow individuals and families over a long period of time. Fortunately, we have population registers from the Belgian commune of Sart that cover almost the entire nineteenth century
 Registres de population, Sart, Communes, Archives de l’État...
suite (Alter and Oris, 2000a). After the census of 1846, the Belgian government introduced a system of population registers designed by the influential statistician Adolphe Quetelet. Each commune was instructed to copy the results of the census into bound volumes and to continuously update these volumes to account for births, deaths, marriages, and migration. Persons who moved from one commune to another were required by law to report these movements at both their origin and destination. Since both taxes and welfare expenses were based on legal domicile, communes had a strong incentive to keep these registers accurately. Although the Belgian population registers were formally designed to record de jure population, we find that they include many de facto movements as well  Since the population registers were intended to be a record...
13 The commune of Sart was unusual in a number of ways. While population registers were introduced in all of Belgium in 1846, there are registers for Sart covering the periods from 1812 to 1843 and 1843 to 1846. This means that we can reconstruct the population of Sart for almost the entire nineteenth century. (Documents referring to the twentieth century are not yet available because of privacy considerations). The earlier registers are not as complete as the later registers, because some migrants were not recorded. However, these omissions are not serious, and we have corrected them as best we can. We focus on Sart because its population registers are unusually complete, but we do not claim that it was a typical Belgian village. In fact, Sart was atypical in a number of important ways.
14 Sart is located in the Ardennes region of Eastern Belgium on the border with Germany. The commune covers an unusually large area, but much of this area is on a plateau covered by a peat bog called the Hautes fagnes. The fagnes are not arable and of limited use for domestic animals. Consequently, the commune has always been sparsely populated. The population lived in several hamlets located on the steep slopes of the plateau, which suffered from poor soil. During the nineteenth century this was recognized as one of the poorest areas in Belgium.
15 Meanwhile, the Industrial Revolution was taking place only a short distance away. In 1798 wool manufacturers brought the English machinist William Cockerill to the city of Verviers, less than 10 kilometers from Sart. Cockerill built the first successful spinning machines outside of England, and the woolen textile industry around Verviers was rapidly transformed from rural proto-industry to mechanical production. A few years later Cockerill and his sons moved from Verviers to the Chateau of Seraing, near the city of Liège about 30 kilometers from Sart, where they built an industrial empire based on iron, coal, and steam engines.
16 Ironically, while industrialization and urbanization were occurring nearby, Sart was de-industrializing. In the eighteenth century a number of forges had been located in Sart to take advantage of charcoal from its abundant forests. These were abandoned by the beginning of the nineteenth century as the forges around Liège converted to coke. Similarly, domestic spinning of wool and even small water-driven mills could not compete with the steam engines in Verviers. In the nineteenth century Sart did participate in the Industrial Revolution by sending timber to the coal mines, but mostly it contributed labor, especially young women who moved to the textile factories and domestic service in urban households. The village grew from about 1,800 inhabitants in 1820 to about 2,500 in 1850, but during the second half of the century out-migration exceeded natural increase. Even agriculture remained poor and difficult until after 1870 when artificial fertilizers were introduced (Vliebergh and Ulen, 1912; see Alter, Oris, Neven, 2000 for a more detailed discussion of the economic and demographic history of Sart).
17 Sart may actually have benefited from its growing isolation in the first half of the nineteenth century. Most of Eastern Belgium suffered an “epidemiological depression”, because the terrible conditions in rapidly growing cities offset any forces for lower mortality (Neven, 1997; Oris, 1998). In contrast, mortality in Sart began to decline shortly after the subsistence crisis that followed the Napoleonic Wars, and improvement was more or less continuous for the rest of the century. Expectation of life increased in the following way:
|Years ||Expectation of life at birth |
|1812-1846 ||39.4 |
|1847-1866 ||40.8 |
|1867-1880 ||45.1 |
|1881-1890 ||46.7 |
|1891-1900 ||55.1 |
18 This trend is visible in the annual Crude Death Rates shown in Figure 1. One indication of Sart’s growing isolation is the changing course of epidemic diseases. In 1834 Sart was severely affected by cholera, but the 1849 epidemic missed the commune entirely. The decline in mortality was not the same across all age groups. Figure 2 shows life tables for the sub-sample used in this analysis, individuals with 4 to 7 siblings (see below). The data is arranged into three cohorts by the date of the mother’s first birth. The largest and earliest decline was in childhood, ages 1 to 15, and mortality in adulthood declined somewhat later. There was almost no change in infant mortality until the end of the nineteenth century.
Fig. 1 - Crude Death Rates in Sart, 1812-1899
Fig. 2 - Age-specific Probabilities of Dying by Year of Mother’s First Birth
19 To place each family in its socio-economic context, we have used property and tax records to construct indicators of wealth. These indicators are based on three different sources. First, from 1818 to1822 the commune of Sart prepared roles for special taxes imposed after an extraordinary period of war, famine, and epidemic. The new tax was based on other taxes paid for real and personal property and business licenses. We have used the first (1818) and last (1822) of these tax lists. Second, in 1843 a new register of the population of Sart was compiled which included the amount of land (in hectares) belonging to each family. These references may include land rented by the household, as well as owned land, but this should not affect our use of these data. Third, a cadastral survey from approximately 1877 (the “Atlas Popp”) includes the value of each property and the owner’s name. Each of these sources has been linked to the population register. To standardize across these different sources we constructed binary variables identifying the top 25 percent of households in each source.
20 Since our analysis is longitudinal, we have constructed three variables that associate our information on wealth with different stages of the life course. We divide the life course into three age groups: childhood (birth to age 15), young adulthood (15 to 40), and later adulthood (40 and older). The variable for each life course stage is the most recent observation in that age group. Unfortunately, in each age group there are many people with missing information, and these missing cases vary across the century. For this reason, we also include a binary variable for persons with missing data, so that they will not affect the comparison between the wealthiest 25 percent and the rest of the population.
21 The model used here includes two kinds of indicators of the family component in mortality
 For a discussion of the use of event history models in historical...
suite. Some measures will be based directly on the experience in each family. For example, we use the number of deaths of siblings as a measure of exposure to disease within families. Deaths are recorded in the population register, and we can construct this variable for each family. However, we also assume that each family influences mortality in ways that we observe only by their consequences. By making some assumptions about the form of this “family effect” we can estimate its overall contribution to differences in mortality.
22 The statistical model used here has the following form:
23 h(a,i,j) is the risk (hazard) of dying at age a for person i in family j,
25 h(a,0,0) is the risk of dying at age a for a standard individual,
26 ß is a vector of coefficients, and
27 x(i,j) is a vector of observed attributes of person i in family j.
28 This model is a “proportional hazards model” of the type described by Cox (1975). It assumes that mortality follows a standard pattern, the “baseline hazard”, which describes changes in the risk of dying at different ages. Observed characteristics cause the risks of dying for each individual to diverge from this standard pattern by the same proportional amount at each age. The estimated coefficients (ß) tell us the effects of each attribute. Estimates of these coefficients are obtained by maximizing Cox’s partial likelihood function, which leaves out the baseline hazard.
29 Differences in mortality among families that are not attributed to observed characteristics (x(i,j)) are associated with w
30 This model gives us two different ways to measure the association between family and mortality. Deaths of siblings are a direct indicator of exposure to disease in childhood. If the means used by “competent” parents to control childhood mortality could be applied to other ages, persons from sibling sets with fewer child deaths should have had lower mortality in adulthood. The random “family effect” (w
31 Models of this kind have previously been used to study clustering of deaths in childhood. In contrast to Das Gupta’s (1990) analysis of Punjab, Guo (1993) found a relatively small “family effect” in data from Guatemala, and Sastry (1997) obtained similar results after controlling for the geographic differences in Brazilian data (Curtis et al., 1993; Zenger, 1993).
32 As far as we know, models of this kind have not been applied to adult mortality, but research on the genetic component of longevity raises similar statistical issues. For example, a study of Danish twins estimates that 20 to 25 percent of the variation in human life span is attributable to genetic variation among individuals (Vaupel, 1998; Yashin and Iachine, 1994). Since siblings share some genetic inheritance, this will contribute to a “family effect” on adult mortality.
33 We have applied models of the kind described above to the life histories of persons with 4 to 7 siblings living in Sart between 1812 and 1900. Information available in the database allows us to link most people to their parents, and we use mother-child links to identify sibling sets. Small and large sibling sets were excluded so that we can use the deaths of siblings in infancy and childhood as explanatory variables
 We also exclude the sibling sets of women whose first birth...
suite. We use binary (“dummy”) variables to indicate which families experienced at least one infant or child death. The risk of a death does increase with the number of siblings, but it is not practical to compute death rates when the number at risk is so low. Limiting the range of family sizes provides some standardization for these differences in risk. We also use separate variables indicating whether a sibling died during infancy or during childhood (ages 1-14), because the factors affecting infant and child deaths can differ substantially (See Table 6 for means of variables).
34 Separate models have been estimated for four age groups: 1-14, 15-29, 30-54, and 55 and older. Preliminary analysis revealed that the association between mortality and previous deaths of siblings was not the same at younger and older ages. Under age 15 the risk of dying was higher for those whose siblings had died, but at older ages mortality was higher among those from families without any child deaths. This changing relationship between the risk of dying and sibling deaths is a violation of the proportional hazards assumption underlying the Cox regression model, which would make our estimation procedures unstable. The simplest solution to this problem is to estimate separate models for different age groups. Not surprisingly, the least successful model appears to be the one for ages 15 and 30 (Table 2), the transitional age group. Mortality is very low at these ages, and this is a time of change in many ways. New experiences, such as new jobs and migration, often carried risks of injury and exposure to disease that are difficult to predict. Nevertheless, the signs and magnitudes of the elements in the model generally fit our expectations and follow the same pattern as other age groups.
35 Tables 1, 2, 3, and 4 summarize the estimated models. Rather than presenting the estimated coefficients (ßs) from the model presented above, we exponentiate these coefficients to get “relative risks”. Relative risks are similar to odds ratios. A relative risk of 2.0 means that a one-unit increase in the covariate (explanatory variable) doubles the risk of death. Similarly, a relative risk of 0.5 indicates that the risk of death decreases by 50 percent, while a relative risk of 1.0 implies that changes in the covariate are not associated with changes in mortality. We believe that relative risks are easier to interpret than coefficients, but it is important to remember that the reference point is 1 not zero.
36 The models include three control variables, which we will describe briefly. Birthyear has been included to capture trends in mortality, which was decreasing in Sart during the nineteenth century. The pace of mortality decline varied by age. It was most rapid in childhood and old age, but the latter is based on a very small range of years. The estimate of 0.95 in Model 1.2 implies that the relative risk of dying in childhood was decreasing by five percent per year. This is a very rapid rate of decline, and it implies that the risk of child mortality fell to less than one hundredth of its starting level over the course of the nineteenth century. Even the estimate of 0.98 for ages 30-54 would have reduced the risk of dying by 86 percent over the course of a century.
37 Family size was negatively associated with mortality at all ages except 15-29, when a larger number of siblings increased the risk of dying. None of these estimates are statistically significant, but the similarity of the results before age 15 and after age 30 is interesting.
38 Sex differences in mortality follow a pattern that we have examined in greater detail elsewhere (Alter, Manfredini, and Nystedt, in preparation). Male mortality was higher in childhood and after age 55, but there was excess female mortality in the prime childbearing years from 30 to 54 (Table 3).
39 Socioeconomic status is measured by the indicators of property ownership described above. The wealthiest quartile of the population had lower risks of dying in every age group, and this association was strongest in childhood and old age. The relative risk of dying for children living in the wealthiest 25 percent of households was only 13 percent of the risk of the average child. At other ages risks of dying were 40 to 50 percent lower in the wealthiest families. Another interesting feature of the models in Tables 3 and 4 is the persistent influence of wealth in childhood and young adulthood. Those who grew up in the top quartile of property-holders were less likely to die at every age, but the effect is particularly strong after age 55. Indeed, it is particularly interesting that wealth during childhood continues to be important even though the models include a measure of wealth at a later age. Our data on wealth are crude, and this may simply mean that two indicators are better than one. On the other hand, it may indicate that childhood conditions had persistent effects that are independent of experiences in adulthood.
40 Both of the family effect variables in Models 1.1 and 1.2 suggest that child mortality tended to cluster within families. If one or more siblings died in infancy, the relative risk of dying was about 30 percent higher than in families with no infant deaths. The high p-values of these estimates imply that they may be due to chance.
41 The unobserved family effect is also large in Model 1.2, and it is close to statistical significance (p-value=.06). To make this part of the model easier to understand, we have estimated the relative risks at the 25th and 75th percentiles of the distribution of the family effect. If we could place all of the families in order by their risks of dying, only 25 percent of the families would have lower risks than the family at the 25th percentile, and only 25 percent would have higher risks than the family at the 75th percentile. In Model 1.2 the estimated relative risks at these points were 0.63 and 1.28 respectively. This means that the healthiest 25 percent of the population was at least 37 percent less likely to die than the average family, and the least healthy 25 percent was 28 percent more likely to die than average. Or, viewed from another perspective, the risk of dying for children in the healthiest quartile of families was less than half the risk in the least healthy quartile. It is also noteworthy that other estimates in the model do not change when we add the random family effect. This implies that the random family effect reflects differences not captured by the sibling mortality variable. The results for ages 15-29 (Table 2) resemble the pattern in childhood, but both the observed and unobserved measures of family effects are much weaker.
42 Above age 30 (Tables 3 and 4) there is a dramatic reversal of the association between the risk of dying and the mortality of siblings. At ages 30 to 54, individuals whose siblings died were almost 50 percent less likely to die. This relationship is much weaker after age 55, but Models 4.1 and 4.2 still imply that mortality was lower for those from a family in which a child died during infancy. Moreover, there is no evidence of an unobserved family effect after age 30 (Models 3.2 and 4.2). Although sickness during childhood may have had some harmful effects, the advantages of acquired immunity appear to have predominated after age 30. In the next section we examine the acquired immunity hypothesis in more detail by looking at mortality during a smallpox epidemic.
43 In Tables 3 and 4 we saw that exposure to disease in childhood, as measured by the deaths of siblings, was associated with lower mortality in adulthood and old age. One explanation for this finding is that survivors of some diseases acquire immunity. If childhood exposure to disease produced immunities, it should be related to mortality from infectious diseases, particularly those known to produce immune responses. Causes of death were not recorded in Belgium until late in the nineteenth century, but it is possible to conduct a simple natural experiment on data from 1871 (Darmon, 1986). In that year a smallpox epidemic spread over most of Europe. The crude death rate in Sart jumped to 39 per thousand, almost double its usual level. This attack was as intense in Sart as in neighboring towns, and it marks the end of Sart’s epidemiological isolation (Neven, 1997).
44 Since smallpox was usually a childhood disease before the advent of vaccination, persons from families with child deaths were more likely to have been exposed to the disease. As a result, they should have been less likely to die during the epidemic of 1871. Table 5 shows exactly this pattern. We examine a sub-sample of 510 persons who were under observation at the beginning of 1871, 23 of whom died during that year. We use logistic regression to test whether those from sibling sets with at least one death in childhood were more likely to survive the year, controlling for age. The estimated odds ratio for any sibling death is.42, indicating that those exposed to more disease in childhood were less than half as likely to die during this epidemic year. Although the sample size is modest, the test of statistical significance indicates that this result is unlikely to occur by chance.
45 This paper has examined evidence of persistent “family effects” in mortality. Like previous studies, we have found strong evidence that child deaths tended to be clustered in some families. The situation in adulthood is more complex, however. Individuals from sibling sets with higher childhood mortality had lower mortality as adults. Controlling for sibling deaths, adult mortality did differ among families, but socio-economic differences explained part of this effect in young adulthood and all of it after age 55. These patterns are suggestive of possible explanations for the “family effect”.
46 During childhood, siblings shared the same environment, and this is reflected in the large family effect under age 15. As they departed from home, however, their environments became increasingly different, and the family effect in nineteenth century Sart weakened with age. It may have had some residual effects in early adulthood, a time when many young adults were still living at home (Capron and Oris, 2000), but the family effect disappeared after age 30. Thus, we do not find any support for the hypothesis that health-promoting behaviors learned in childhood were beneficial in adulthood. On the contrary, our evidence points to the acquired immunity hypothesis.
47 Higher exposure to disease was an important risk factor in childhood mortality, but it had benefits in later life. Individuals from sibling sets with higher child mortality tended to survive longer in adulthood. Our analysis of the 1871 smallpox epidemic suggests that immunities acquired during childhood played an important role in nineteenth-century mortality. It also raises interesting questions about the dynamics of mortality in a period when epidemics were receding. Clearly, vaccination reduced death rates, but it also tended to create pools of people at greater risk. It was not immediately recognized that the immunity conferred by vaccination deteriorated over time. In addition, as epidemics became less common, more people could reach adulthood without encountering the disease (Pitkanen et al.,1989). Of course, the advantage of acquired immunity disappears when the likelihood of ever encountering the disease becomes very small. Thus, the benefits of surviving exposure to diseases in childhood depend upon the specific pathogens that a person is likely to encounter as a child and as an adult.
48 It is difficult to see the operation of genetic differences in the evidence presented here. Since deaths at younger ages were primarily from infectious diseases, we would expect shared genetic traits to be most apparent in old age, when degenerative diseases are more important. However, we observe a stronger family effect at younger than older ages. Using different methods, a number of previous historical studies (Cournil, 1996; Gavrilova et al., 1998) have found genetic effects on survival, but it is not too surprising that the effect is not noticeable here. On average, siblings only share one quarter of their genetic inheritance, so it will be a smaller part of the family effect in these data than in studies of twins or parents and children.
49 Our analysis also revealed a strong effect of socio-economic status at all ages. Living in a better-off household as a child reduced mortality after age 55 even when we controlled for wealth in early adulthood. We cannot rule out deficiencies in our data, but we may be seeing the persistent effects of conditions in childhood. In a related study, we have shown that height at age 20 was strongly related to old-age mortality and that men from wealthier households were significantly taller (Alter, Oris, 2000b; Fogel, 1993; Waaler, 1984). This implies that poverty in childhood had long-lasting physiological effects.
|Model 1.1 ||Model 1.2 |
|Relative risk ||p-value ||Relative risk ||p-value |
|Birthyear ||0.96 ||0.00 ||0.95 ||0.00 |
|Number of siblings ||0.88 ||0.11 ||0.87 ||0.18 |
|Female ||1.13 ||0.48 ||1.09 ||0.62 |
|Any infant deaths ||1.27 ||0.20 ||1.32 ||0.22 |
|Any child deaths |
|Property above 75 percentile in childhood ||0.15 ||0.00 ||0.13 ||0.00 |
|Property unknown in childhood ||0.84 ||0.33 ||0.84 ||0.38 |
|Relative risk of family effect: |
|at 25th percentile ||0.63 |
|at 75th percentile ||1.28 |
|Variance of family effect ||0.52 |
|p-value of family effect ||0.06 |
|Observations ||918 ||918 |
|Time at risk ||10180.9 ||10180.9 |
|Deaths ||267 ||267 |
|Likelihood ratio test ||69.40 ||174.00 |
|Degrees of freedom ||6.00 ||52.50 |
|p-value of Likelihood ratio test ||0.00 ||0.00 |
|Model 2.1 ||Model 2.2 |
|Relative risk ||p-value ||Relative risk ||p-value |
|Birthyear ||0.99 ||0.12 ||0.99 ||0.12 |
|Siblings at age 15 ||1.15 ||0.22 ||1.15 ||0.22 |
|Female ||1.00 ||0.99 ||1.00 ||0.99 |
|Any infant deaths ||1.14 ||0.66 ||1.14 ||0.66 |
|Any child deaths ||1.09 ||0.76 ||1.09 ||0.76 |
|Property above 75 percentile in childhood ||0.59 ||0.25 ||0.59 ||0.25 |
|Property unknown in childhood ||0.78 ||0.46 ||0.78 ||0.46 |
|Property above 75 percentile in young adulthood ||0.56 ||0.12 ||0.56 ||0.12 |
|Property unknown in young adulthood ||0.53 ||0.11 ||0.53 ||0.11 |
|Relative risk of family effect: |
|at 25th percentile ||0.96 |
|at 75th percentile ||1.04 |
|Variance of family effect ||0.00 |
|p-value of family effect ||0.35 |
|Observations ||839 ||839 |
|Time at risk ||10072.2 ||10072.2 |
|Deaths ||161 ||161 |
|Likelihood ratio test ||12.80 ||13.30 |
|Degrees of freedom ||9.00 ||9.23 |
|p-value of Likelihood ratio test ||0.17 ||0.16 |
|Model 3.1 ||Model 3.2 |
|Relative risk ||p-value ||Relative risk ||p-value |
|Birthyear ||0.98 ||0.08 ||0.98 ||0.08 |
|Siblings at age 15 ||0.90 ||0.39 ||0.90 ||0.39 |
|Female ||1.20 ||0.48 ||1.20 ||0.48 |
|Any infant deaths ||0.46 ||0.02 ||0.46 ||0.02 |
|Any child deaths ||0.53 ||0.05 ||0.53 ||0.05 |
|Property above 75 percentile in childhood ||0.59 ||0.20 ||0.59 ||0.20 |
|Property unknown in childhood ||0.92 ||0.80 ||0.92 ||0.80 |
|Property above 75 percentile in young adulthood ||0.62 ||0.13 ||0.62 ||0.13 |
|Property unknown in young adulthood ||0.85 ||0.69 ||0.85 ||0.69 |
|Relative risk of family effect: |
|at 25th percentile ||1.00 |
|at 75th percentile ||1.00 |
|Variance of family effect ||0.00 |
|p-value of family effect ||0.93 |
|Observations ||469 ||469 |
|Time at risk ||6369.3 ||6369.3 |
|Deaths ||61 ||61 |
|Likelihood ratio test ||15.50 ||15.50 |
|Degrees of freedom ||9.00 ||9.00 |
|p-value of Likelihood ratio test ||0.08 ||0.08 |
|Model 4.1 ||Model 4.2 |
|Relative risk ||p-value ||Relative risk ||p-value |
|Birthyear ||0.95 ||0.23 ||0.95 ||0.23 |
|Siblings at age 15 ||0.84 ||0.38 ||0.84 ||0.38 |
|Female ||0.56 ||0.13 ||0.56 ||0.13 |
|Any infant deaths ||0.70 ||0.45 ||0.69 ||0.45 |
|Any child deaths ||1.17 ||0.74 ||1.17 ||0.74 |
|Property above 75 percentile in childhood ||0.36 ||0.06 ||0.36 ||0.06 |
|Property unknown in childhood ||2.74 ||0.05 ||2.75 ||0.05 |
|Property above 75 percentile in young adulthood ||0.61 ||0.28 ||0.61 ||0.28 |
|Property unknown in young adulthood ||3.82 ||0.06 ||3.83 ||0.06 |
|Relative risk of family effect: |
|at 25th percentile ||1.00 |
|at 75th percentile ||1.00 |
|Variance of family effect ||0.00 |
|p-value of family effect ||0.94 |
|Observations ||113 ||113 |
|Time at risk ||1115.8 ||1115.8 |
|Deaths ||38 ||38 |
|Likelihood ratio test ||20.20 ||20.20 |
|Degrees of freedom ||9.00 ||9.00 |
|p-value of Likelihood ratio test ||0.02 ||0.02 |
|Covariates ||Odds ratio ||p-value |
|Age ||1.07 ||0.32 |
|Age squared ||1.00 ||0.21 |
|Any sibling deaths ||0.42 ||0.04 |
|Overall p-value ||0.10 |
|Observations ||510 |
|Deaths ||23 |
|Ages 1-14 ||Ages 15-29 ||Ages 30-54 ||Ages 55+ |
|Birthyear ||1849.8 ||1849.8 ||1845.1 ||1849.8 |
|Number of siblings (ages 1-14)/ |
Siblings at age 15 (Ages 15+)
|6.4 ||5.3 ||5.2 ||5.3 |
|Female ||0.473 ||0.473 ||0.448 ||0.473 |
|Any infant deaths ||0.347 ||0.347 ||0.322 ||0.347 |
|Any child deaths ||0.131 ||0.486 ||0.550 ||0.486 |
|Property above 75 percentile in childhood ||0.131 ||0.131 ||0.164 ||0.131 |
|Property unknown in childhood ||0.476 ||0.476 ||0.488 ||0.476 |
|Property above 75 percentile in young adulthood ||0.160 ||0.228 ||0.160 |
|Property unknown in young adulthood ||0.408 ||0.296 ||0.408 |
Alter, George (1998), « L'Event History Analysis en démographie historique. Difficultés et perspectives », Annales de Démographie Historique, 2, 25-35.
Alter, George, Capron, Catherine, Neven, Muriel, Oris, Michel (forthcoming), “When Dad Died: Household Economy and Family Culture in Nineteenth Century East Belgium”, in When Dad Died. Individuals and Families Coping with Family Stress, Renzo Derosas, Michel Oris (eds.).
Alter, George, Neven, Muriel, Oris, Michel (2000), “Individuals, Households and Communities Facing Economic Stresses: a Comparison of Two Rural Areas in xixth Century Belgium”, 185-210, in Family Structures, Demography and Population. A Comparison of Societies in Asia and Europe, Muriel Neven and Catherine Capron (eds.), Liège, Laboratoire de Démographie.
Alter, George, Oris, Michel (2000a), “Mortality and Economic Stress: Individual and Household Responses in a Nineteenth-Century Belgian Village”, 335-370, in Population and Economy: From Hunger to Modern Economic Growth, Tommy Bengtsson and Osamu Saito (eds.), Oxford, Oxford University Press.
Alter, George, Oris, Michel (2000b), “Early Life Conditions and the Decline of Adult Mortality in the Belgian Ardennes, 1812-1890”, presented at Population Association of America, Los Angeles, March 23-25.
Alter, George, Manfredini, Matteo, Nystedt, Paul et al. (in preparation), “Gender and Mortality”, to be published in Families, the Economy and Death in Past Times: Relationships between Economic Conditions, Household Context, and Mortality in Europe and Asia before the Twentieth Century, Tommy Bengtsson and Cameron Campbell (eds.).
Bengtsson, Tommy, Saito, Osamu (2000), Population and Economy: From Hunger to Modern Economic Growth, Oxford, Oxford University Press.
Bideau, Alain, Biraben, Jean-Noël, Dupâquier, Jacques (1988), « La mortalité de 1800 à 1914 », 279-298, in Histoire de la Population Française, t. 3, Paris, PUF.
Capron, Catherine, Oris, Michel (2000), « Ruptures de cohabitation entre parents et enfants. L’expérience des campagnes et des villes du Pays de Liège au xixe siècle », 229-269, in De l'usage des seuils. Structures par âge et âges de la vie, éd. par Alain Bideau, Patrice Bourdelais, Jacques Légaré, Paris, Société de Démographie historique.
Cournil, Amandine (1996), Mise en évidence d’une composante familiale de la longévité. L’exemple de la population de la Valserine (Ain) aux xviiie et xixe siècles, Rapport Technique dans le cadre du D.E.A. « Analyse et modélisation des systèmes biologiques », Université Claude Bernard-Lyon I.
Darmon, Pierre (1986), La longue traque de la variole. Les pionniers de la médecine préventive, Paris, Librairie académique Perrin.
Derosas, Renzo (Forthcoming), “La fortuna di nascere ebrei. Fattori culturali nei differenziali di mortalità infantile” [The luck of being Jew. Cultural factors in differential infant mortality], in Studi in onore di Marino Berengo, Milano, Franco Angeli.
Dupâquier, Jacques (1982), « Implications et conséquences de l'allongement de la vie humaine depuis le xviiie siècle », 215-220, in Le vieillissement. Implications et conséquences de l'allongement de la vie humaine depuis le xviiie siècle, éd. par Arthur E. Imhöf, Lyon, Presses universitaires de Lyon.
Elo, Irma T., Preston, Samuel H. (1997), “Racial and Ethnic Differences in Mortality at Older Ages”, 10-42, in Racial and ethnic differences in the Health of Older Americans, Linda G. Martin and Beth J. Soldo (eds.), Washington, D.C., National Academy Press.
Gavrilova, Natalia S., Gavrilov, Leonid A., Evdokushkina, Galina N., Semyonova, Victoria G., Gavrilova, Anna L., Evdokushkina, Natalia N., Kushnareva, Yulia E., Kroutko, Vyacheslav N., Andreyev, Alexander Yu (1998), “Evolution, Mutations, and Human Longevity: European Royal and Noble Families”, Human Biology, 70, 4, 799-804.
Guo, Guang, Rodriguez, German (1992), “Estimating a Multivariate Proportional Hazards Model for Data Using the EM Algorithm, with an Application to Child Survival in Guatemala”, Journal of the American Statistical Association, 87, 420, 969-76.
Hardy, Anne (1993), The Epidemic Streets. Infectious Diseases and the Rise of Preventive Medicine, 1856-1900, Oxford, Oxford University Press.
Havelange, Carl (1990), Les Figures de la Guérison (xviiie-xixe siècles), Paris, Société d’Edition « Les Belles Lettres », Bibliothèque de la Faculté de Philosophie et Lettres de l’Université de Liège, fasc. CCLV.
Hélin, Étienne (1960), La genèse et la disparition des isolats. Remarques méthodo-logiques et sondages préliminaires dans la Province de Liège, unpublished thesis of the University of Liege (Social Sciences).
Knodel, John (1988), Demographic Behavior in the Past: a Study of Fourthteen German Village Populations in the xviiith and xixth Centuries, Cambridge, Cambridge University Press.
Lecuyer, Bernard P. (1986), « Hygiène en France avant Pasteur, 1750-1850 », 67-139, in Pasteur et la révolution pastorienne, éd. par Claire Salomon-Bayet, Paris, Payot.
Lunn, Peter G (1991), “Nutrition, Immunity, and Infection”, 131-145, in The Decline of Mortality in Europe, Roger Schofield, David-S. Reher, and Alain Bideau (eds.), Oxford, Clarendon Press.
Lynch, Katherine A., Greenhouse, Joel B. (1994), “Risk Factors for Infant Mortality in Nineteenth-Century Sweden”, Population Studies, 48, 1.
Neven, Muriel (1997), “Epidemiology of Town and Countryside. Mortality and Causes of Death in East Belgium, 1850-1910”, Revue Belge d’Histoire Contemporaine, 1-2, 39-82.
Oris, Michel (1998), « Mortalité, industrialisation et urbanisation au xixe siècle. Quelques résultats des recherches liégeoises », 289-322, in Dix essais sur la démographie urbaine de la Wallonie au xixe siècle, éd. par Claude Desama et Michel Oris, Bruxelles, Crédit communal de Belgique.
Pozzi, Lucia, Robles Gonzáles, Elena (1996), “L’analisi della mortalità infantile negli anni della transizione: una riflessione sull’esperienza italiana e spagnola”, 147-171, in Secondo Convegno dei Giovani Studiosi dei Problemi di Popolazione, Studi di popolazione. Temi di ricerca nuova, Rome, SIDES.
Preston, Samuel H. (1985), “Resources, Knowledge and Child Mortality: A Comparison of the US in the Late Nineteenth Century and Developing Countries Today”, in International Population Conference, Florence, 41, 373-86.
Preston, Samuel H., Haines, Michael R. (1991), Fatal Years: Child mortality in Late Nineteenth-Century America, Princeton, New Jersey, Princeton University Press.
Reher, David S., Schofield, Roger (1991), “The Decline of Mortality in Europe”, 1-17, The Decline of Mortality in Europe, Roger Schofield et al. (eds.), Oxford, Clarendon Press.
Riley, James C (2001), Rising Life Expectancy: A Global History, Cambridge, Cambridge University Press.
Riley, James C., Alter, George (1996), “The Sick and the Well: Adult Health in Britain during the Health Transition”, Health Transition Review, 6, Suppl., 19-44.
Sastry, Narayan (1997a), “A Nested Frailty Model for Survival Data, with an Application to the Study of Child Survival in Northeast Brazil”, Journal of the American Statistical Association, 92, 438, 426-35.
Sköld, Peter (1996), The Two Faces of Smallpox, Umeå, Sweden, Umeå University.
Thornton, Patricia, Olson, Sherry (1992), « Familles montréalaises du xixe siècle : trois cultures, trois trajectoires », Cahiers québecois de Démographie, 21, 2, 51-75.
Vaupel, James W., Carey, James W., Kaare, Christensen, et al. (1998), “Biodemographic Trajectories of Longevity”, Science, 280, 855-60.
Vliebergh, Émile, Ulens, Rob (1912), L’Ardenne. La population agricole au xixe siècle, Bruxelles, Librarie Albert Dewit.
Waaler, Hans Th. (1984), “Height, Weight and Mortality: The Norwegian Experience”, Gruppe for helsetjenesteforskning, Rapport nr. 4, Oslo: Statens institutt for folkehelse.
Woodbury, Robert Morse (1926), Infant Mortality and its Causes, with an Appendix on the Trend of Maternal Mortality Rates in the United States, Baltimore, The Williams & Wilkins Company.
Yashin, Anatoli I., Iachine, Ivan A. (1994), “Genetics of Human Longevity: Danish Twin Data Applied to a Demographical Question”, Population Studies of Aging #11, Center for Health and Social Policy, Odense University.
 Physicians in the Province of Liège did not begin to promote systematic re-vaccination until after 1864. The weak quality of the vaccine, from human origin, remained problematic until the adoption of cow-pox vaccine in the late 1880s (Oris, 1995, 989-990). French experience was similar (Darmon, 1986, especially p. 358).
 Registres de population, Sart, Communes, Archives de l’État à Liège.
 Since the population registers were intended to be a record of the living population, infants who died shortly after birth were often not recorded. We have added information about these children from the birth and death registers, which are generally considered complete.
 For a discussion of the use of event history models in historical studies, see Alter, 1998.
 We also exclude the sibling sets of women whose first birth occurred before 1812, and those who were under observation less than 15 years after their last birth.
Les différences de mortalité entre les familles fournissent un indice de l'importance des comportements inobservés conditionnant à la santé. Par exemple, si la plus faible mortalité était due à des attitudes personnelles apprises au cours de l'enfance, elle devrait se manifester jusqu'aux âges élevés. Dans cet article sont utilisées des données provenant d'une commune belge, au xixe siècle, afin de préciser les différences de mortalité entre les familles de deux manières. Tout d'abord, un indice de l'exposition des enfants à la maladie est construit en comptant le nombre de décès à moins de 15 ans dans chaque famille. Ensuite, l'effet global des différences entre les familles est calculé grâce à un modèle aléatoire qui permet d'estimer la variance de « l'effet famille ». Les deux mesures mettent en évidence un effet familial fort pendant l'enfance, qui diminue après 15 ans puis disparaît au-delà de 55 ans. En outre, à une période encore dominée par les maladies infectieuses, ceux qui survivent aux maladies contractées au cours de l'enfance acquièrent des immunités qui les rendent plus résistants par la suite.
Differences in mortality among families give us clues about the importance of unobserved health-related behaviors. For example, if lower mortality was due to types of personal behavior learned in childhood, it should carry over to mortality at older ages. In this paper we use records from a nineteenth-century Belgian community to look at differences at mortality differences among families in two ways. First, we construct a direct measure of exposure to disease in childhood by counting the number of children in each family that died before age 15. Second, we calculate the overall effect of inter-family differences by using a "random effect" model that estimates the variance of the "family effect". Both of these measures show a strong family effect in childhood, but this effect diminishes after age 15 and disappears after age 55. Moreover, in a period still dominated by infectious diseases, those who survived diseases in childhood acquired immunities that helped them in later life.
PLAN DE L'ARTICLE
- Health-promoting behaviors and the mortality of siblings
- Acquired immunity and heterogeneous frailty
- Data and setting
- Statistical Models
POUR CITER CET ARTICLE
George Alter et al. « The Family and Mortality: A Case Study from Rural Belgium », Annales de démographie historique 1/2001 ( no 101), p. 11-31.
URL : www.cairn.info/revue-annales-de-demographie-historique-2001-1-page-11.htm.