1 – The study population

13We used three criteria to select the study population. First, it was necessary to limit our selection to individuals born between 1625 and 1704. This avoids analytical problems associated with right censoring because the PRDH had not yet finished recording most deaths beyond 1800 at the time of the study. The range allowed us to observe the survival patterns of the early French Canadian individuals to approximately 100 years of age. To avoid bias, we selected individuals who were born to mothers known to have had all their children before 1705. Individuals not meeting this criterion were removed because their inclusion would have biased the estimates toward the mortality patterns of lower birth-order individuals.

14Second, we used age 50 as the minimum age of inclusion. The minimum age was imposed because infant mortality tends to be under-registered, which biases upwards the life expectancy at birth (e.g. e₀ = 43.72 years for these individuals). Mortality risks from infectious diseases and from external causes such as maternal and accidental deaths (Olshansky and Ault, 1986) may also mask significant effects at the older ages, as some external causes are age-invariant and “unrelated” to old-age mortality (Vaupel, 1988).

15Third, we selected individuals conditional upon having at least one other sibling surviving past age 50. Applying the above criteria gave us a base population of 7,448 individuals. The addition of other familial controls further restricted the size of the study population because complete information on the age at death of the parents and spouses was not always available. For the bivariate models, we have 77.0% complete information on spouses, 83.2% on fathers and 88.5% on mothers. When we included spousal and parental ages at death simultaneously (Table 2, Models 1-5), we ended up with 61.4% complete information, or an overall study population of 4,560 (2,295 males and 2,265 females).

Table 2

Factors influencing the risk of death for individuals surviving past age 50 (Quebec, seventeenth and eighteenth centuries). of the Cox proportional hazard models

Covariates Males Females Multivariate models (N = 2,295) Multivariate models (N = 2,265) Bivariate models Model 1 Model 2 Model 3 Model 4 Model 5 Bivariate models Model 1 Model 2 Model 3 Model 4 Model 5 Birth group x geographic location       (1625-1679) x West ref. ref. ref. ref. ref. ref. ref. ref. ref. ref. (1680-1694) x West 1.1 1.07 1.09 1.1 1.07 1.1 1.1 1.09 1.1 1.1 [0.080] [0.080] [0.080] [0.080] [0.080] [0.077] [0.077] [0.077] [0.077] [0.077] (1695-1704) x West 1.26** 1.22* 1.25** 1.24** 1.20* 1.04 1.04 1.03 1.05 1.04 [0.080] [0.080] [0.080] [0.080] [0.080] [0.078] [0.078] [0.078] [0.078] [0.078] (1625-1679) x East 1.28** 1.28** 1.27** 1.27** 1.26** 1.21* 1.20* 1.18* 1.22* 1.18* [0.084] [0.084] [0.084] [0.084] [0.084] [0.078] [0.078] [0.079] [0.078] [0.079] (1680-1694) x East 1.46*** 1.40*** 1.44*** 1.44*** 1.38*** 1.40*** 1.37*** 1.38*** 1.40*** 1.35*** [0.082] [0.082] [0.082] [0.082] [0.082] [0.078] [0.078] [0.078] [0.078] [0.079] (1695-1704) x East 1.38*** 1.35*** 1.36*** 1.36*** 1.32*** 1.30*** 1.31*** 1.29** 1.30** 1.28** [0.083] [0.083] [0.083] [0.083] [0.083] [0.08] [0.08] [0.08] [0.08] [0.08] Residence area Rural ref. ref. ref. ref. ref. ref. ref. ref. ref. ref. Urban 1.19** 1.19** 1.19** 1.20*** 1.19*** 1.17*** 1.17** 1.16** 1.17*** 1.16** [0.055] [0.055] [0.055] [0.055] [0.055] [0.050] [0.050] [0.051] [0.050] [0.051]

Covariates Males Females Multivariate models (N = 2,295) Multivariate models (N = 2,265) Bivariate models Model 1 Model 2 Model 3 Model 4 Model 5 Bivariate models Model 1 Model 2 Model 3 Model 4 Model 5 Age at death of siblings () 0.981*** 0.978*** 0.979*** 0.987*** 0.984*** 0.984*** (N = 3,515) [0.003]  [0.003] (N = 3,913)  [0.003] [0.003] of spouse 0.993*** 0.993*** 0.994*** 0.995** 0.995** 0.995** (N = 2,999) [0.002] [0.002] (N = 2,756) [0.002] [0.002] of mother 0.996*** 0.996** 0.997** 0.997**  0.998 0.998 (N = 3,152)  [0.001] [0.001] (N = 3,537)  [0.001] [0.001] of father 0.999  1.00 1.00 0.996**  0.999 0.999 (N = 2,944)  [0.002] [0.002] (N = 3,233)  [0.002] [0.002] Model fit (?2) 35.62*** 84.26*** 55.09*** 44.57*** 107.46*** 36.89*** 60.76*** 43.58*** 40.08*** 70.94*** Note: p <.001***, p <.01**, p <.05*; numbers between brackets are asymptotic standard errors (+/–). Reading: For categorical variables, 1.19 indicates that those residing in urban areas had a 19% higher risk of death than those living in rural areas. For continuous variables such as the siblings’ age at death 0.979 indicates a 2.1% [(0.979-1) x 100] lower risk of death for each year increase in the mean survival of the sibship. Source: Registre de Population of the Programme de recherche en démographie historique (PRDH), University of Montreal.

Factors influencing the risk of death for individuals surviving past age 50 (Quebec, seventeenth and eighteenth centuries). of the Cox proportional hazard models

2 – Age at death of siblings

16The core variable of the study is the survival influence of siblings. Like parents and their offspring, siblings share 50% of their genes (Gagnon et al., 2005), and this may create correlated survival times among them. Siblings, however, are also closer in age to each other than to their parents. This implies that they share more similar social and environmental conditions than with their parents and these circumstances probably have a lifelong effect on survival outcomes. In historical Quebec, the family worked together to acquire a large resource base and to establish itself in the newly forming colony (Bouchard, 1996; Gagnon and Heyer 2001). We hypothesize that, as a result, a mutually supportive and productive sibship would be beneficial to individual survival over the life course. Were those individuals who had longer-lived siblings also more likely to live longer themselves?

17In order to answer that question, we followed two different strategies.

18We first computed the average age at death of the siblings of an individual surviving past age 50 as:

19

20where S_i represents the age at death of sibling i and n is the number of siblings, excluding the individual in focus. The formula gives each individual a unique average age at death of his or her siblings who survived past age 50 [3]. We constrained the minimum span of at age 50 to avoid interactions with time at the younger ages (or violation of proportionality) as a result of maternal mortality and other external causes of death [4].

21Second, we randomly assigned only one of the sibling’s ages at death to each individual and repeated this procedure a 100 times to obtain an overall average of the coefficient. The first method, based on the average or , was used for efficiency and predictive purposes (i.e., the average age at death of the sibship is a better predictor of survival than the age at death of a single sibling). The second method, based on random replicates (i.e. a bootstrap), was used to obtain an estimate of the sibling effect that is not affected by sibship size and to ease the comparison with the other familial effects described below. Our method is similar to a bootstrap, where the procedure approximates the sampling distribution of the coefficients and standard errors by drawing repeated random samples (resampling) with replacement.

3 – Other familial and environmental variables

22We controlled for other familial effects by including maternal, paternal and spousal ages at death. In order to preserve a maximum number of cases in the analysis, we did not truncate these covariates at age 50, as we did for siblings [5]. Maternal and paternal ages at death were entered as continuous covariates in the model to analyse the possible combination of genetic and/or shared environmental effects and their influence on the lifespan of the individual. Additionally, spousal age at death was included as a continuous predictor to control for cohabitation or shared social and environmental effects between marriage partners.

23To control for period and regional effects, we included urban/rural residence, geographic location and birth year of the individual in all of the models. Individuals who lived in Quebec city, Trois-Rivières and Montreal were classified as urban residents. The rest of the population were designated as rural dwellers and were used as the reference category. The geographical areas were divided into two districts (east and west) based on the divisions by Gagnon and Heyer (2001). The west includes Montreal, Trois-Rivières and the surrounding rural parishes, whereas the east includes Quebec City and the eastern parishes of the colony. Birth year was collapsed into 3 groups that consisted of individuals born between: 1) 1625 and 1679; 2) 1680 and 1694; 3) 1695 and 1704. The particular coding captures the effects of living through times when there was a high risk of death from infectious diseases. Since the temporal effects were conditional upon the area, we included a cross-product term of these covariates (interaction term). Individuals born between 1625 and 1679 and who lived in the western district were designated as the reference category. Table 1 summarizes the coding of the nominal covariates and gives the number of males and females in each category for the models in Table 2.

Table 1

Description of the categorical variables included in models 1 to 5

Covariates Males Females Residence Rural 1,864 1,718 Urban 431 547 Birth Group 1625-1679 609 684 1680-1694 848 842 1695-1704 838 739 Area West 1,131 1,130 East 1,164 1,135 Number 2,295 2,265 Source: Registre de Population of the Programme de recherche en démographie historique (PRDH), University of Montreal.

Description of the categorical variables included in models 1 to 5

4 – Cox proportional hazard models

24A series of Cox regression models were fit, to assess whether the sibling predictors had any influence on the individual’s survival time. The Cox regression model expresses a transformation of the hazard as a linear function of the predictors. A continuous hazard function is a rate with no upper bound (?) and thus the logarithm of the hazard is treated as the outcome variable (Singer and Willet, 2003, p. 514):

The log hazard log h(t_i, X) equals the baseline function log h₀(t) plus a weighted linear combination of predictors ? that measure the effect of the covariates on log h(t_i, X), which is nil when the covariates equal 0. The main assumptions of the Cox proportional hazards model are 1) a log-linear relationship between the covariates and the underlying hazard function, and 2) a multiplicative relationship between the underlying hazard function and the log-linear function of the covariates (Blossfeld et al., 1989, Courgeau and Lelièvre, 1992). It is assumed that the hazard functions of any two individuals have parallel age (time) patterns (Namboodiri and Suchindran, 1987; Elandt-Johnson and Johnson, 1980). All of the covariates included in our models appeared to meet the proportionality assumption [6].

1 – Kaplan-Meier survival curves

26Figures 1 and 2 show the Kaplan-Meier survival curves from age 50 of individuals born between 1625 and 1704 (2,295 males and 2,265 females, respectively). These individuals were divided into two groups based on whether their siblings’ average age at death was above or below 69.9 years (median of ). The survival curves clearly demonstrate that individuals with longer-lived siblings had a better survival experience. A Log Rank test confirms that the difference between the two groups was highly significant (p < 0.001). Males with longer-lived siblings had a median survival time of 70.6 years, almost 2.5 years more than the other group (68.2 years). Similarly, females with longer-lived siblings had a median survival time of 71.9 years versus a median of 69.7 years for the other group.

Figure 1

Kaplan-Meier survival functions for males with the average age at death of their siblings (Quebec, 17^th and 18^th centuries)

Kaplan-Meier survival functions for males with the average age at death of their siblings (Quebec, 17^th and 18^th centuries)

Figure 2

Kaplan-Meier survival functions for females with the average age at death of their siblings (Quebec, 17^th and 18^th centuries)

Kaplan-Meier survival functions for females with the average age at death of their siblings (Quebec, 17^th and 18^th centuries)

2 – Bivariate and multivariate Cox proportional hazards models

27Table 2 gives the proportional effect for all variables, obtained as the hazard ratios for the bivariate and multivariate models (1-5) that include controls for period and familial effects. The bivariate hazard ratios are all significant, with the exception of the paternal-son association. Models 1-4 include each of the familial covariates while controlling for period and regional effects, and Model 5 gives all covariates entered simultaneously. Models 1-4 are nested with model 5 and will be described accordingly. Model 5 had the best overall fit, as can be seen by the larger ?² difference from the base model (107.46 for males and 70.94 for females; p < 0.001 in both cases). All hazard ratios reported below were taken from Model 5. Overall, the Cox models provided a better fit to the male data than to that of their female counterparts.

28The birth group × area interaction was significant for both sexes, as well as the rural/urban mortality differential. These coefficients are stable across all models. In general, mortality was higher in all of the birth groups from the eastern area of the colony. Men living in eastern area had a mortality risk that was successively 26.0%, 38.0% and 32.0% higher than that of the earliest western inhabitants [7]. The corresponding figures for females were 18%, 35% and 28% (p < 0.05, p < 0.001 and p < 0.01). As time went by, mortality also increased in the west, but apparently only for males. Males born between 1695 and 1704 from the western area had a 20% higher mortality hazard than the first generation. The corresponding estimates for females are not significant. Additionally, men living in the urban centers (Montreal, Trois-Rivières and Quebec City) had a 19.0% higher risk than those living in rural areas, whereas women had a 16.0% higher hazard in urban areas.

29The presence of longer-lived siblings on average proves to be the strongest predictor of old-age survival for both men and women (Models 2 and 5). The average age at death of the sibship was highly significant among both sexes (p < 0.001), but the effect was stronger for males. An annual increase in gave males a 2.1% lower risk and females a 1.6% lower risk of death. Over a decade, this would cumulate to approximately an 18.9% lower risk among men and a 14.8% lower risk of death for women (e^(-0.016×10) = 0.811 and (e^(-0.016×10) = 0.852).

30Models 3 and 5 illustrate the effects of including the spousal age at death. The spousal age at death was significant for both males and females, although the relative effect was stronger for males. Increasing the age at death of a wife by one year meant a 0.6% lower mortality risk for males (or a 5.8% lower risk for a 10 year increase). Individuals with wives who survived to age 87.3, for example, had an 11.3% lower mortality risk than those with wives that died at age 67.3 years. For females, an annual increase in their husbands’ age at death was associated with a reduction in mortality risk of 0.5% or of 4.9% over a 10-year span.

31For both males and females, the paternal age at death was not a significant predictor in Models 4 and 5. On the other hand, the maternal age at death had a significant effect on male survival, although this effect was small. Males had a 0.3% lower risk of death for each additional year that their mother survived, holding the other covariates constant (p < 0.01) [8]. For instance, a man whose mother died at age 84.2 (i.e. 20 years after the average) had a 5.8% lower hazard ratio than one whose mother died at age 64.2 (the average). There was no significant maternal-daughter association in the multivariate models.

32Since the data contain correlated observations or the possibility of temporal dependence among groups of individuals (e.g. siblings), we also ran the same models with robust variance estimation (results not shown). This procedure involves relaxing the temporal independence assumption by accounting for the clustering of observations. Under those conditions, the standard errors and significance levels of the coefficients remained remarkably similar to the more “traditional” estimates reported in Table 2. To further ensure the consistency of our results, we also ran simulation models, using the same set of controls as in Model 5. One individual was randomly selected from each family, where each of the samples contained 1,294 males and 1,282 females. After a 100 trials, the average hazard ratios for were 0.984 (with asymptotic standard errors [ase] = 0.004) for males and 0.985 ([ase] = 0.004) for females, respectively. These estimates are very close to the hazard ratios obtained by including all siblings (0.979 for males and 0.984 for females; see Table 2), which indicates that the correlations among siblings do not affect the results substantially.

33There is, however, a possibility that the effect size of is larger than that of the parental and spousal effects simply because the average is a more efficient measure than a single numeric value. For instance, fathers, mothers and spouses are single individuals for whom we have only one measured age at death, and such a single measurement is more subject to random variation than the average age at death of the siblings. We thus ran simulations to check for the potential artifact by randomly selecting one sibling’s age at death for each individual and then took the averages of the hazard ratios after 100 replicates (Table 3). All of the other covariates present in Table 2 were also included as controls in the simulations. We did not report the controls in Table 3 because their effect sizes remained constant over the 100 replicates. As expected, the effect size associated with the (randomly chosen) siblings’ age at death decreased. Nevertheless, it remained larger than the spousal and parental effects. The average hazard ratio was 0.990 for males (this represents a decrease of 1% for each additional year in the sibling’s age at death), and 0.995 for females.

Table 3

Variation of the effect of sibling’s age at death on risk of death for individuals surviving past age 50 (Quebec 17^th and 18^th centuries) with randomly selected sibling (Cox model)

 Covariate Males Females Randomly selected sibling 0.990*** 0.995** (N = 2,295) (N = 2,265) Randomly selected brother living in the same region 0.988*** 0.993* (N = 1,300) (N = 1,183) Randomly selected sister living in the same region 0.992** 0.995* (N = 1,325) (N = 1,259) Note: Results are adjusted for residence (urban/rural, east/west), birth cohort, spousal, and parental ages at death as defined in Table 2. We derived a Wald test statistic for each estimate by dividing the average parameter estimate by the average standard error over the 100 replicates. p <.001***, p <.01**, p <.05* Reading: the ratio of .0.990 indicates that the individual had a 1.0% lower risk of death for each year increase in the age at death of the sibship. Source: Registre de Population of the Programme de recherche en démographie historique (PRDH), University of Montreal.

Variation of the effect of sibling’s age at death on risk of death for individuals surviving past age 50 (Quebec 17^th and 18^th centuries) with randomly selected sibling (Cox model)

34In order to test for the effects of shared environment and of social support among siblings, we also present the hazard ratios pertaining to the ages at death of brothers and sisters residing in the same region as ego [9] (Table 3). Here, we used more detailed geographic subdivisions consisting of 10 regions with each containing 10 to 15 parishes as defined by Gagnon and Heyer (2001). Because of data selection (i.e. not every individual had at least one brother or one sister living in the same region) the sample sizes are smaller than in the previous models. Nevertheless, it is apparent that the association is stronger among brother-brother pairs than brother-sister or sister-sister pairs. For males, each additional year in their brother’s age at death is associated with 1.2% lower mortality risk, while an additional year in a sister’s age at death is associated with an 0.8% lower risk. For females, the corresponding reductions in risk are 0.7% and 0.5%, respectively. Hence, both males and females benefited from the close proximity of a longer-lived sibling (in particular a brother), although the magnitude of the effect was slightly larger for males. The differences are not overwhelming, but in general – and this was also apparent in Table 2 – a sibling’s age at death is a better predictor of his/her brother’s survival than of his/her sister’s survival.

35Using the parish instead of the region of residence yielded similar results, although these results were less significant because of the reduced sample size. We also compared sex-specific sibling pairs without reference to the region of residence (not shown here). The resulting effect sizes were all smaller than those obtained when taking the region of residence into consideration. Hence, sex alone is not the key factor because siblings’ associations in age at death were larger when they were both living in the same region. Additionally, we tested whether the number of siblings co-residing in the same region was associated with a reduced mortality risk. The coefficients for this variable were not significant and thus not reported (not shown here). Lastly, we re-estimated the sibling coefficients by randomly assigning to each individual a sibling who lived in a different region. The hazard ratio for males was 0.992 (p < 0.05), which represents an effect size slightly lower than the one obtained when not considering the region of residence (i.e. hazard ratio = 0.990; see Table 3). The hazard ratio for females was 0.996.

1 – Epidemics and the pioneering nature of settlement

37We attempted first to isolate a major environmental influence on survival, that of epidemics. Infectious diseases, generally considered as “childhood diseases” in historical Europe, had a large impact on adults in New France (Desjardins, 1996). Inhabitants were distributed over a large and sparsely populated area and this gave them a lower chance of being exposed to immunogenic viruses during infancy and childhood. As adults, they often had little or no acquired immunity when an outbreak swept through the colony. The variation, however, was largely conditional upon geographic location, justifying the inclusion of the birth group × area interaction term.

38Most epidemics originated in Quebec City (eastern area), the main port of entry to New France, and from there spread to the rest of the colony. The first major epidemic occurred around 1687 (typhoid fever) and others, such as smallpox and measles, began appearing with greater frequency throughout much of the eighteenth century. Consequently, our results showed that all of the birth groups from the eastern region had a higher risk of death. These individuals were exposed to more frequent epidemics and lived in more crowded conditions (especially in the urban centres). As for the western residents, only males belonging to the most recent cohorts (1695-1704) had a significantly different mortality risk than their predecessors. In the latter periods, expansion and increased settlement (or contact density) of the western frontier allowed viruses to spread more rapidly.

2 – Familial factors of longevity: heredity and shared traits?

39Amid varying environmental conditions, some individuals could have benefited from stronger vitality inherited from their parents. Thus, we initially suspected that the parent-offspring associations would parallel the other Quebec studies, as well as their European counterparts. In this regard, our bivariate results were largely consistent with those obtained by Desjardins and Charbonneau (1990) and by Blackburn et al. (2004). Both of those studies concluded to a genuine, but relatively small heredity effect. In our study, the introduction of controls was responsible for a substantial dampening of the parental effects, which suggests that the previous bivariate associations were spurious to some extent. Only the mother-son association remained significant. This parent-offspring association was indeed identified as the strongest among the four possible pairs in Desjardins and Charbonneau’s study (1990). However, the father-son pair was not significant in any of our models, nor was it significant in the Blackburn et al. (2004) study. Our study used a sample selection procedure different from those used in the two previous studies. The selection criteria along with the incomplete information on parents and spouses (necessary for the inclusion of controls) resulted in a smaller study population, particularly in the multivariate models, and this could partly explain the differences between our results and those of the previous studies.

40We believe that, significant or not, the strength of parent-offspring association in longevity cannot be overwhelmingly strong in historical data. Other studies support this argument. Concerning maternal transmission, two French studies (Cournil et al., 2000; Bocquet-Appel and Jakobi, 1990) reported no maternal-offspring association. Meanwhile, a mother-offspring association only appeared in the later cohorts of the British Peerage (Westendorp and Kirkwood, 2001). The latter study covered a time when maternal mortality and other external causes were having less impact on female survival (i.e. during the industrial revolution). In our study, periodic epidemics and a high risk of maternal mortality may partly account for the absence of parent-daughter associations. Women’s chances for survival might have been largely determined by their reproductive history, which may explain the lower fit to the models pertaining to females, as compared with models pertaining to males.

41Survival to old age was apparently independent of parental longevity, but it was strongly dependent on the survival of siblings. The average age at death of the siblings was the strongest predictor of an individual’s age at death after age 50. In agreement with the New England, Okinawa, Utah and Icelandic studies (Perls et al., 2002a; Willcox et al., 2006; Kerber et al., 2001; Gudmundsson et al., 2000), individuals with longer-lived siblings had a lower mortality hazard themselves. A remarkable consistency of the sibship coefficients was observed in each of our models. The sibling effect suggests several interpretations. The constant lower mortality risk of individuals with longer-lived siblings (Figures 1 and 2) could favour an interpretation of positive genetic traits endowed at birth (Perls et al., 2002b). This genetic variation would include a lack of predisposition to degenerative diseases, stronger disease resistance or a slower rate of senescence (Perls and Terry, 2003; Perls et al., 2002a; Vaupel, 1988). Like parent-offspring pairs, sibling pairs may share additive genetic (or polygenic) variation through the 50% of genes they have in common over the whole genome. But as explained earlier, siblings may share dominance interactions as well as additive variations. In the Danish twin study, models based on dominance components provided better fit to the data than models based on additive genetic variance (Herskind et al., 1996; McGue et al., 1993).

3 – Shared genetic inheritance or shared life conditions? Large differences between the sexes

42Most of the pre-eminence of the sibling effect over the parental effect may, however, come from the fact that, to a large extent, siblings share the same environmental and social conditions throughout their lives. The large age differences between parents and children (typically around 25 to 30 years apart) would allow period and cohort effects to accumulate and possibly make environmental conditions quite different between generations. In colonial New France, mortality risks may have further clustered within sibships because siblings congregated in mutually supportive groups. The family was considered a “collective and egalitarian unit” and its members tended to migrate together and establish farms within close proximity (Bouchard, 1994). Bouchard (1992) indicates that sons emigrated to settle newly opened land and the family helped with the initial clearing. As pionniers accapareurs (“monopolizing pioneers”), siblings would cooperate in taking over large stretches of land to establish themselves and their descendants (Matthieu et al., 1992; Gagnon and Heyer, 2001). This characteristic allowed family members to remain close to one another, and could explain the stronger association between siblings residing in the same region in comparison with other pairs of siblings (Table 3).

43In all instances, associations between males were larger than the associations between females, whether or not the region of residence was considered. The fairly constant difference could be partly explained by the sharing, among brothers, of critical variants on the Y-chromosome (which cannot be shared within brother-sister or sister-sister pairs). Although further analyses will be needed to test this possibility, we believe that the difference between males and females, in this regard, is mostly due to the fact that females’ correlations were largely distorted by fertility-related covariates, as written above. There is a possibility that the robustness of the females made their survival partly independent from their family members. It must be noted that these women survived through the childbearing ages under a high (natural) fertility regime, with an average of 9.2 children (Charbonneau 1975). Any woman who survived numerous and frequent pregnancies (under conditions of lowered immunity) may be seen as a “robust” individual. Her chances for survival to older ages were probably already higher, regardless of the conditions of her family members. In fact, in our sample, the average age at death of a woman surviving past age 50 was 70.9 years, as compared to 69.2 years for males. The lower fit of the models to the female data, combined with their higher rate of survival indicates that unobserved mechanisms are in operation.

44Like siblings, spouses were also exposed to similar life conditions because of their closeness in age and similar social backgrounds (e.g. assortative mating). Marriage and cohabitation would further expose the couple to the same adult lifestyle, familial norms and material resources. As a result, we found that both males and females benefited from longer-lived spouses and these results are similar to the ones obtained by Blackburn et al. (2004) and Westendorp and Kirkwood (2001). For females, the spousal effect was similar in magnitude to the effect associated with the age at death of a randomly chosen sibling (Tables 2 and 3). Overall, the accumulation of resources among males would benefit their spouses while wives would influence male survival through the provision of domestic duties (Greer, 1997). Strong kinship ties and mutual aid would buffer against hardships over the adult life course. Nevertheless, the exposure to similar circumstances during early life would probably remain with the siblings over the life course regardless of whether they resided in the same region or not. We continually found a significant sibling effect and believe it is partly the result of the lifestyles, habits, values and norms learned in early life. For example, even if an individual moved to a different region in adulthood, his/her survival could still be associated with that of his/her siblings based on the similar conditions experienced during childhood. Simply, these traits and early life exposures probably remain throughout life to influence the actions and behaviors of an ageing sibship.

45Sufficient evidence was gathered here to show that longevity is a complex phenotype that cannot be studied without explicit reference to historical, social and environmental circumstances. By comparison, the genetic influence appears relatively modest. It essentially rests on the indication that the association among siblings (whether they lived in the same region or not) was persistent and generally larger than the association between spouses. It would be presumptuous to argue that the familial transmission described in other studies was always an artifact. But we note that the introduction of a few variables capturing the effects of historical changes and environmental variations (period effects, region of residence) was sufficient to blur the associations between parents and offspring. Larger sample sizes and more complex methods will be needed to disentangle the factors of longevity in historical Quebec. Recent research has insisted on the need “to move beyond descriptive heritability and to aim for precise specification of genetic and environmental mechanisms as well as their interaction” (Crow and Johnson, 2005, p. 7). We also feel that future avenues explicitly addressing this complexity are most promising, particularly those avenues that will go beyond the loose specification of the environmental influences.

46* * *