Males’ Later-Life Mortality Consequences of Coresidence With Paternal Grandparents: Evidence From Northeast China, 1789–1909

Abstract

In this study, we investigate the effect of early-life coresidence with paternal grandparents on male mortality risks in adulthood and older age in northeast China from 1789 to 1909. Despite growing interest in the influence of grandparents on child outcomes, few studies have examined the effect of coresidence with grandparents in early life on mortality in later life. We find that coresidence with paternal grandmothers in childhood is associated with higher mortality risks for males in adulthood. This may reflect the long-term effects of conflicts between mothers and their mothers-in-law. These results suggest that in extended families, patterns of coresidence in childhood may have long-term consequences for mortality, above and beyond the effects of common environmental and genetic factors, even when effects on childhood mortality are not readily apparent.

Introduction

Family and household context in childhood are of great significance for various aspects of later-life well-being in both developed (Elo et al. 2014; Galobardes et al. 2006; Hayward and Gorman 2004; Preston et al. 1998) and developing countries (Cunningham et al. 2010; Engle and Breaux 1998; Evans and Miguel 2007; Morrell et al. 2003). One contextual feature of obvious interest is coresidence with different types of kin. Studies of effects of childhood coresidence with parents already show that their presence or absence in early life may have long-term consequences: children who lived with a single parent tend to have a higher risk of dying in their later life than children who lived with both parents (Campbell and Lee 2009; Elo et al. 2014; Hayward and Gorman 2004). Of course, these effects are at least partly indirect, working through adult socioeconomic status (SES) and household context, which in turn determine health and mortality outcomes (Elo et al. 2014; Martikainen et al. 2009).

In light of such findings, accumulated evidence on the effects of grandparents’ presence on child survival suggests the possibility that it may also influence mortality in later life. Evidence for the immediate beneficial effects of presence of maternal grandmothers is the most consistent. Studies in countries such as Germany, Japan, India, and Canada have found that the presence of maternal grandmothers is associated with higher chances of child survival, possibly because grandmothers help provide childcare (for a summary, see Sear and Mace 2008). In patrilineal societies such as China, children have much more contact with paternal than maternal grandparents. However, few studies have explored effects on child survival of coresidence with paternal grandparents in patrilineal societies. Among these studies, results are inconsistent (for a summary, see Sear and Coall 2011; Sear and Mace 2008). Relevant studies of eighteenth and nineteenth century China have suggested that coresidence with a paternal grandfather increased child mortality (Campbell and Lee 1996, 2002).

Understanding the long-term influence of childhood coresidence with grandparents is important. In most historical and some contemporary societies, the norm is for grandparents to live with one or more adult children and their grandchildren. This is certainly true for the site of our study, northeast China, as well as other societies in Asia (Chen et al. 2014; Knodel and Ofstedal 2002; Logan et al. 1998; Takagi and Silverstein 2006). Although some studies have argued that multigenerational coresidence is on the decline, new evidence suggests that it is once again increasing in some developing countries (Ruggles and Heggeness 2008). Even where multigenerational households have become rare, a sizable fraction of the contemporary elderly population may have spent time in one as children.

Our analysis also offers a new perspective on grandparent effects on later-life outcomes. As part of the growing attention to multigenerational demographic and stratification processes (Mare 2011), there is new interest in the implications of coresidence or other interactions with grandparents in childhood on outcomes later in life. However, most work so far has focused on measurement of grandparent influences on socioeconomic attainment (Chan and Boliver 2013) or determinants of long-term reproductive success as measured by total numbers of descendants two or more generations later (Hawkes 2004).

We examine the later-life mortality consequences for males of childhood coresidence with paternal grandparents in preindustrial northeast China. We not only examine effects of coresidence with grandparents in childhood, but we also seek to assess the roles of three channels: (1) direct effects on child health that also have long-term consequences, (2) selection effects, and (3) indirect effects. Although our exclusive focus on paternal grandparents is due to data limitations described later, it is also the case that in historical China and other historical and contemporary societies characterized by patrilocal marriage (i.e., in which wives join their husband’s household), children are much more likely to coreside with paternal grandparents than with maternal grandparents. The restriction to studying male mortality in adulthood and old age, meanwhile, reflects other limitations of the data that we describe shortly.

For our analysis, we use the China Multigenerational Panel Dataset, Liaoning (CMGPD-LN), which describes approximately 250,000 individuals living in 600 largely rural communities in northeast China from 1789 to 1909. We apply discrete-time event-history analysis (Allison 1984) to measure male mortality risks in adulthood and older ages according to childhood coresidence with a paternal grandparent, with controls for other features of household context in childhood. To control for unobserved characteristics of the household and community that may have influenced the survival of grandparents and grandchildren, we also estimate models with father fixed effects.

This study has at least two distinguishing features. To our knowledge, it is the first study of the mortality consequences in adulthood and old age of childhood coresidence with paternal grandparents for any society in which such coresidence was the norm. Examination of the long-term consequences of presence of a paternal grandparent in childhood is important in societies such as historical China in which married adults coresided with surviving elderly parents, usually the husband’s. Second, results from this preindustrial and largely agrarian setting may offer insights into the legacy of related processes in contemporary societies, where patriarchal, multigenerational households were also the norm, and the population was largely rural when today’s elderly were still children. In such societies, the health and mortality of the elderly now may reflect long-term influences of childhood coresidence with grandparents like those reported here. If data emerge that allow for studies similar to ours for such societies, comparison of the results will yield insight into how specific features of local context condition the relationship between coresidence with paternal grandparents and later-life health outcomes.

Background

Coresidence With Grandparents and Child Health

Previous studies have suggested a number of pathways by which coresidence with a paternal grandfather may have influenced children’s health. The left side of Fig. 1 summarizes these pathways. Coresidence with a grandfather might benefit child health by improving economic well-being or increasing family stability. The former refers to the possibility that in preindustrial societies, grandfathers contributed to the economic well-being of the household by providing labor (Tymicki 2004, 2009). The latter refers to the possibility that in a patrilocal society like China (where wives join their husband’s household at time of marriage), the death of a paternal grandfather might lead to conflicts within the household over inheritance or other issues, and that these conflicts could adversely affect child health (Tymicki 2004, 2009). Conversely, an adverse effect of coresidence with a grandfather might arise from conflict between junior and senior generations over resources. Children and the elderly all depend on resources generated by working-age adults. In a patriarchal society like China, males in the senior generation take precedence over junior males. As a result, children who coresided with their paternal grandfather may have been allocated fewer resources than those whose paternal grandfather had already passed away (Campbell and Lee 2004).

Results from previous studies suggest a number of mechanisms by which coresidence with a paternal grandmother may have had an immediate beneficial or adverse effect on infant and child health, depending on the social and cultural context. The left side of Fig. 2 summarizes possible pathways by which the presence of a paternal grandmother may influence infant and child health. Many studies have emphasized the beneficial direct effects on children of care from a paternal grandmother or the indirect effects arising from help she provides to her daughter-in-law (Beise 2005; Cunningham et al. 2010; Derosas 2002; Griffiths et al. 2001; Kemkes-Grottenthaler 2005; Leonetti et al. 2005). Fewer studies have reported a negative effect of coresidence with paternal grandmother on infant and child mortality. Explanations for these negative effects focus on conflicts with in-laws (Beise 2002; Voland and Beise 2002; Willfuhr 2009).

Empirical studies on the effect of coresidence with paternal grandparents on mortality risks in childhood have yielded conflicting results. The review of kin effects on child survival by Sear and Coall (2011) included a number of studies that considered effects of paternal grandparents. Results vary according to the choice of age range. Common age ranges include 0–2, 0–5, 0–10 and 0–15. Generally speaking, for studies focusing on ages 0–2, age 0–10, or age 0–15, coresidence with a paternal grandfather increases mortality in childhood, and coresidence with a paternal grandmother reduces it (Jamison et al. 2002; Kemkes-Grottenthaler 2005; Tsuya and Kurosu 2002, 2004). For those studies focusing on ages 0–5, coresidence with a paternal grandfather reduces mortality, and coresidence with a paternal grandmother has mixed effects (Beise 2002, 2005; Tymicki 2004, 2009; Voland and Beise 2002; Willfuhr 2009).

Early-Life Influence on Adult Health

A large body of literature has documented the association between childhood conditions and health in later life. Adversity in early life—whether as the result of competition with coresident paternal grandfathers over resources, family instability following the death of a paternal grandfather, or in-law conflicts between mothers and coresident paternal grandmothers—might increase mortality later in life through scarring or reduce it through selection, as summarized in the right sides of Figs. 1 and 2. On the one hand, adverse conditions in early life that have an immediate effect on mortality risks may permanently impair the health of those who survive, leading to higher death rates later in life (Elo and Preston 1992; Kuh and Wadsworth 1993; Mosley and Gray 1993)—a physiological scarring effect. On the other hand, various forms of selection effects may lead children who survived a risky childhood to experience lower death rates later in life. Individuals who survived adversity in childhood may have started out more robust (Alter et al. 2001; Elo and Preston 1992; Preston et al. 1998).

Childhood coresidence with paternal grandparents might also have indirect effects on health and mortality later in life. Such coresidence might directly affect the diet, lifestyle, socioeconomic attainment, or other health-related behaviors of children in adulthood or old age, and thereby influence their health and mortality (Jendrek 1993; Zeng and Xie 2014). Furthermore, as suggested in the right sides of Figs. 1 and 2, resource competition or family stability associated with childhood coresidence with grandparents could affect SES in adulthood, and thus affect their health and mortality.

Much of the discussion of possible mechanisms of the long-term influences of childhood context has focused on the potential role of experiences during critical periods of development. According to the critical period model, shocks that occur when specific organ systems are developing may impair health in later life because they have long-term effects on the functioning of those systems (Ben-Shlomo and Kuh 2002). Conditions in utero have received particular attention because different organs develop at specific times during pregnancy (Hayward and Gorman 2004). Indeed, this is the basis of the original fetal origins hypothesis (Barker 1992). Conditions in the first years of life are also important because infancy and early childhood are crucial periods of growth and development.

Hypotheses

Based on our reading of this literature, we specify a set of hypotheses corresponding to contrasting scenarios for mechanisms by which coresidence with grandparents may have affected health in later life. We consider paternal grandfathers and paternal grandmothers separately, with one pair of hypotheses (Hypotheses 1a and 1b) for effects of paternal grandfathers and another pair (Hypotheses 2a and 2b) for paternal grandmothers. We also consider a hypothesis (Hypothesis 3) to account for the possibility that any observed benefit of coresidence in later life might be the production of selection processes in childhood.

• Hypothesis 1a (H1a): Coresidence with a paternal grandfather in childhood increases mortality later in life because discrimination in favor of senior males in the allocation of household resources led to physiological scarring of junior males.

• Hypothesis 1b (H1b): Childhood coresidence with a paternal grandfather lowers mortality later in life because the labor they contributed improved household economic standing, and his presence promoted family stability (Tymicki 2004, 2009).

• Hypothesis 2a (H2a): Childhood coresidence with a paternal grandmother increases mortality later in life because of physiological scarring caused by in-law conflicts between paternal grandmothers and mothers in childhood.

• Hypothesis 2b (H2b): Childhood coresidence with a paternal grandmother lowers mortality later in life because of the assistance that they provided to the mother.

• Hypothesis 3 (H3): Childhood coresidence with paternal grandfathers or grandmothers lowers mortality later in life because the higher infant and child mortality produced by resource competition with them (Campbell and Lee 1996, 2002) selected out the least-healthy children and/or left surviving children immune to the diseases responsible for the higher mortality.

Data

We make use of data from China Multi-Generational Panel Dataset, Liaoning (CMGPD-LN), 1749–1909 (Lee and Campbell 2011). The CMGPD-LN consists of 1,513,357 triennial observations of 266,091 individuals who lived in rural northeast China between the late eighteenth and early twentieth centuries. It is organized by household, kin group, village, administrative district, and region. The data provide extensive details on household relationship, age, name(s) and name changes, occupation, marriage, emigration, geographic location, and SES as well as information on fertility and mortality. Importantly, the CMGPD-LN also includes a constructed outcome variable specifying whether an individual died in the next three years. This is specifically intended for use as a dependent variable in a discrete-time event-history analysis of mortality. Because the data are publicly available via ICPSR and are already well documented (Dong et al. 2015; Lee and Campbell 1997; Lee et al. 2010), our discussion here focuses on features directly relevant to the analysis.¹

The data have three features especially relevant to the needs of this analysis. First, in contrast with most contemporary longitudinal surveys of the elderly, the CMGPD-LN records early-life and adult circumstances prospectively rather than relying on retrospective reports by respondents. The records in the CMGPD-LN not only follow males across their life course but also link them to their parents and paternal grandparents. At every point in time, details on household context are available. This addresses a major difficulty in studying the determinants of health over the life course in a multigenerational perspective: the scarcity of prospective data that not only follow individuals from childhood to old age but also follow families and households over multiple generations. Second, the suitability of the CMGPD-LN for analysis of mortality has already been established, having been used in several published studies of mortality (Campbell and Lee 1996, 2002, 2004) and at least two studies of early-life influences on adult and old-age mortality (Campbell and Lee 2009; Dong and Lee 2014). Properties of the data relevant to the study of mortality are known, and limitations have been identified and accounted for. Third, the population was closed. Entries into and exits from the population were extremely rare; and when they did occur, they were annotated. In contrast with most other historical population databases, after people left their origin village for another one, they were still recorded. We do not need to worry about differential loss to follow-up among individuals according to their mortality risks (Lee and Campbell 1997).

The CMGPD-LN also has limitations relevant to this analysis. First, records of many children who died in early childhood were missing, so we are unable to reliably calculate mortality rates in infancy or early childhood as a proxy for disease exposure in early life (Forsdahl 2002; Leon and Smith 2000). Along these lines, direct measures of exposure to disease that are central to theories about influences of early-life conditions on health in later life (Ben-Shlomo and Kuh 2002) are not available.

Second, the CMGPD-LN allows study of childhood contextual influences on only later-life mortality for males because at present, it is not possible to follow a woman from her natal household to her husband’s household after she marries. The problem is that the available data do not allow for records of married women to be linked back to their records as daughters in their natal households. Although the ideal study would examine both men and women, we are not able to conduct such a study, and we avoid generalizing from our results for males.

Third, our variables for age at loss of grandparent or other kin are based on the age at which they were last recorded in a register as alive rather than a recorded age at death. Records in the CMGPD-LN specify that a death has occurred in the last three years, but they do not provide the exact date of death. If a death was recorded in a register that is missing, we know only the year in which the individual was last recorded as alive in the available data. To have a consistent measure, we rely on age in sui at the last live appearance of a grandparent,² which will tend to underestimate the age at death of grandparents.

We apply a variety of restrictions to the observations in the data set to account for other limitations of the CMGPD-LN. We include data from 1789–1909 because registers before 1789 did not distinguish separate residential households within larger administrative units.³ We restrict the data to observations of individuals who were between 1 and 75 sui and for whom another observation is available three years later. The latter restriction ensures that we consider only time intervals during which a death in the next three years could have been recorded in a surviving register. We also exclude individuals whom we are unable to link to parents or grandparents, and those cases for which values for mother’s age at the child’s birth are implausible.⁴ Our final sample consists of 406,335 observations describing the life courses of 85,677 males.

Methods

We apply discrete-time event-history analysis (Allison 1984) to examine the roles of the channels through which coresidence with paternal grandparents in childhood may influence later-life mortality risks. Discrete-time event-history analysis via logistic regression is an appropriate method in light of the structure of the data. Previous studies of mortality in the CMGPD-LN have all used discrete-time event-history analysis because, as noted earlier, the data specify only the three-year interval in which a death occurred rather than the actual date of death (Campbell and Lee 1996, 2001, 2002, 2004). Discrete-time event-history analysis is specifically designed for such data, when the interval in which an event occurred is known but the actual date is not (Allison 1984). Although complementary log-log regression should yield coefficients that are directly comparable to those produced by proportional hazards models (Long 1997), our previous experiments with both forms of regressions indicate that the results are nearly identical.

To control for unobserved community or family characteristics that simultaneously influence grandparent survival and grandchild mortality in childhood or later in life, in addition to estimating models with controls for observed characteristics of the household, we also estimate models that include fixed effects for fathers. In such models, we can rule out the possibility that an observed association is the product of the influence of unobserved characteristics of the community and household on the health of the grandparent and the grandchildren.

To adjudicate among our hypotheses, we run separate analyses for the effect of coresidence with paternal grandfather in early life on mortality in childhood, adulthood, and old age. If scarring is important (H1a and H2a), childhood coresidence with a grandparent should increase mortality in childhood, adulthood, and old age. If selection is important, childhood coresidence with a grandfather should raise child mortality but lower adult and old-age mortality (H3). If labor contributions and family stability play a role (H1b), childhood coresidence with a paternal grandfather should lower child mortality as well as mortality in later life. If coresidence with a paternal grandmother is beneficial because of the assistance that she provided to the mother (H2b), it should be associated with lower mortality risks later in life.

We estimate logistic regressions with and without fixed effects of grandfather for three separate age groups: boys aged 1–15 sui, adult males aged 16–55 sui, and elderly males aged 56–75 sui. These ranges are intended to correspond roughly to 0–15, 15–55, and 55–75 Western years of age. Within each age group, Model 1 includes the indicators of grandparents’ coresidence in the year of birth; and controls for age, region, year, and other characteristics, such as household context and SES. To account for unobserved characteristics of the household that affect mortality risks, Model 2 adds a fixed effects of fathers and eliminates variables that will be the same among brothers, such as time-invariant characteristics of their parents (Allison and Christakis 2006; Hoffman and Duncan 1988). Results for effects of coresidence with the paternal grandfather reflect comparisons between brothers according to whether their grandfather was still alive when they were children.

Our first set of analyses focuses on the association between mortality risks and coresidence with grandparents in or around the year of birth. Here, coresidence with grandparents in the year of birth serves as a proxy for whether the child coresided with them in infancy and early childhood. Because of how the CMGPD-LN is structured, the variable for coresidence with grandparents in year of birth will actually reflect coresidence with that grandparent from birth up to the time of the next triennial household register. Typically, this will be sometime in infancy or early childhood.

In the rural society covered by the CMGPD-LN, adult children almost always coresided with the father’s parents. Thus, if a paternal grandparent was alive, a child almost always lived with him or her. An adverse effect of coresidence with a grandparent around time of birth would imply that grandchildren competed with the grandfather for household resources (H1a) or suffered from in-law conflicts between the grandmother and the mother (H2a). Conversely, a beneficial effect would imply that the grandfather improved household environments (H1b), or the grandmother assisted parents with care (H2b) or selection effects (H3).

To assess the influence of duration of coresidence with grandparents, we use a critical period model that recognizes the potential influence of exposure during specific periods in childhood. We conduct a parallel analysis to determine whether the influence of coresidence on child mortality varies by the age of the child, and whether duration of childhood coresidence conditions mortality outcomes in adulthood and old age. Each set of analyses consists of two models: one with and one without fixed effects of fathers. These examine the immediate effects on mortality of coresidence at specific ages in childhood and the longer-term effects of the duration of coresidence with the grandparent. In the historical Chinese context, because adult couples coreside with the father’s parents, grandchildren almost always live with their surviving paternal grandparents. In this context, a child’s age when a grandparent died doubles as a measure of the duration of their period of coresidence with that grandparent. Because most theories about critical periods emphasize the importance of conditions in utero or infancy, we expect that coresidence with grandparents at early ages will be most important and that continued coresidence at later ages will have only minor additional effects.

We also examine whether the effect of coresidence with a grandparent around time of birth was conditioned by the presence or absence of a parent of the same sex. We hypothesize that any adverse effect of coresidence with a grandmother (H2a) would be weaker for a child who lost their mother because there was no possibility for conflict between her and her mother-in-law. Any beneficial effects of coresidence with a grandmother (H2b) might be magnified because the grandmother might substitute for the mother in providing care or ensure that another family member would provide care. For this, we include variables indicating parental survival status and longevity as well as interaction terms between grandparents’ and parents’ survival status around the time of birth. Similarly, we expect that for children who lost their father around the time they were born, the adverse effect of a grandfather would be weaker (H1a), or the beneficial effect of a grandfather would be stronger (H1b), because a grandfather might partially substitute for the father in supplying labor for the household.

Measures

The key right-side variables in the current analysis are coresidence with paternal grandfather and grandmother at time of birth. We first create two dummy variables to indicate the coresidence of paternal grandfather and grandmother around the time of the child’s birth. We treat “survival of grandparents” and “coresidence with grandparents” as interchangeable because if paternal grandparents were alive, children almost always coresided with them (Lee and Campbell 1997). To construct this measure, we compare each individual’s calculated year of birth to the register year in which their grandmother or grandfather was last recorded as alive.⁵ If the year of last observation for the grandparent is equal to or greater than the child’s calculated year of birth, the grandparent is considered to be coresident at the time of birth.⁶

We use the child’s age when their grandparent was last recorded as alive as a proxy for their age at the time of their grandparent’s death, and the duration of their childhood coresidence with that grandparent. For the analysis of mortality of children under age 16 sui, we test the immediate effects on mortality of coresidence at specific ages in childhood by interacting current age group with whether the paternal grandfather and paternal grandmother are currently alive. For the analyses of adult and old-age mortality, we divide individuals into three categories according to their age when their paternal grandfather or grandmother was last recorded: 1 to 5 sui, 6 to 10 sui, or 11 sui and older to test the longer-term effects of the duration of coresidence with the grandparent. The reference category includes men whose paternal grandfather or grandmother died before they were born.

Basic controls include age, year, and region. As noted earlier, ages were recorded in sui in original registers. In the analysis, age in sui is entered as a quadratic polynomial. Region is entered as dummy variables, indicating north, central, south central, or south Liaoning. Mortality levels vary dramatically across these regions. Mortality was lowest in south and south central Liaoning, higher in north Liaoning, and highest in the central region, which was also the most densely settled. We control for year because previous studies have suggested a trend of declining infant and child mortality in the population during the nineteenth century (e.g., Campbell and Lee 2004).

We also include controls for features of household context that previous studies have shown to be associated with mortality in historical China (e.g., Campbell and Lee 2009) and that may plausibly be correlated with the presence of grandparents. We capture an individual’s childhood SES by creating dummy variables indicating whether their father held a salaried official position or had a diminutive name. We capture their status in adulthood in old age by creating dummy variables for whether the individual held a position. Having a salaried official position is a sign of being local elite, and having a diminutive name in historical China was generally associated with low status.⁷

Controls for household context include maternal birth age, paternal birth order, number of present brothers, birth interval, presence of parents, and grandparents’ longevity. Maternal birth age, short preceding birth interval, paternal birth order, number of present brothers, and presence of parents in childhood are found to significantly affect both child and adult mortality (Campbell and Lee 2009). Maternal age at birth is measured using two indicators. One identifies individuals who were born when their mother was 20 sui or younger, and the other identifies individuals who were born when their mother was 35 sui or older. Number of brothers is calculated directly from the data. A short preceding birth interval is defined as a maternal birth interval from zero to two years. Dummy variables are included to indicate whether the individual’s father or mother was currently alive. We also include an indicator of grandparents’ exceptional longevity as an alternate approach to account for correlated mortality risks as a result of common genes or environment. To capture those cases where grandparents lived to especially old ages, we measure grandparents’ longevity using two dummy variables that indicate whether a grandfather or a grandmother lived to more than 60 sui.

We also control for marital status as a potential intervening variable in adult and old-age mortality. Previous studies have also shown that married males have lower mortality than unmarried males in many countries (e.g., Hu and Goldman 1990). A previous exploration of CMGPD-LN shows that parental survival influenced male marriage chances: men whose parents were still alive were more likely to marry (Chen et al. 2014). Family SES also influences marriage chances, with men from higher status families more likely to marry early. Marital status may therefore have been a proxy for otherwise unobserved aspects of SES in historical northeast China. Marital status is recoded as a dummy variable indicating whether an individual was currently married.

Table 1 summarizes the variables in the analysis divided by age group. For our key independent variables, the grandfather was alive at the time of the subject’s birth for approximately 38 % of the observations, and the grandmother was alive for approximately 56 % of the observations. These percentages are very similar across all three age groups. This mitigates concerns of sample selection bias stemming from the possibility that coresidence with grandparents at time of birth might affect the chances of survival to adulthood or old age.

Descriptive Results

Descriptive results are consistent with H1b or H2b: individuals who coresided with a paternal grandfather at time of birth have higher overall survival chances in adulthood and old age, while coresidence with a grandmother at time of birth is beneficial in adulthood, not old age. Figure 3 presents the Kaplan-Meier smoothed hazard estimates in adulthood and old age for individuals who survived to at least age 16 sui, according to whether they coresided with a grandfather or grandmother at birth. The mortality hazards for people who coresided with a grandfather at birth are lower than for those who did not, particularly during adulthood. Until age 60, the mortality hazards for people who coresided with a grandmother at birth are slightly lower than for those who did not. At that point, there is a crossover, and the mortality hazard for those who coresided with a grandmother at birth becomes higher than for those who did not. A log rank test shows strong, significant differences between the hazards by grandfathers’ survival status at birth (χ²(1) = 26.69, p = .000). Overall, there are fewer deaths among persons whose grandfathers were alive at birth than those whose grandfathers were not. The effect of grandmothers’ survival status at birth is weakly significant (χ²(1) = 2.88, p = .090).

Descriptive results for the effects of coresidence with a grandparent around time of birth if the parent of the same sex passed away are inconclusive. Figures 4 and 5 present the relevant results. There are no significant differences between the hazard curves by coresidence with a grandfather (χ²(1) = 0.35, p = .553) or grandmother (χ²(1) = 0.22, p = .636) at birth, as indicated by log rank tests. However, among people whose parents were still alive at their time of birth, there are fewer deaths among people who coresided with a grandfather at birth than among those whose grandfathers died before birth (χ²(1) = 22.74, p = .000). Although visual inspection of the Kaplan-Meier smoothed hazard estimates in Figs. 4 and 5 may suggest larger differences according to the survival status of a grandparent when men lost a parent of the same sex around time of birth, these differences are not statistically significant.

Event-History Analysis

Presence of Paternal Grandparents at Time of Birth

To assess whether selection processes existed that could have produced a pattern in which coresidence increased child mortality but reduced mortality later in life (H3), we begin with an assessment of effects of coresidence on child mortality. Analyzing an earlier, smaller version of the CMGPD-LN, Campbell and Lee (1996, 2004) reported adverse effects of presence of a grandfather, which they attributed to resource competition with a dependent patriarch. They examined overall effects of grandparent coresidence on mortality of children between ages 2 and 15 but did not distinguish effects by age group. Results in Model 1 of Table 2 confirm the adverse effect of coresidence with paternal grandfather on child mortality. However, after the introduction of fixed effects for father in Model 2, associations are no longer statistically significant. The apparent effect of coresidence with a grandfather in Model 1 may be an artifact of association of child mortality with unobserved community or household characteristics that had an opposite effect on grandfather’s mortality. No significant effects are found for grandmothers in any model.

Presence of a paternal grandmother at the time of birth has clear adverse effects on adult male mortality. Models 1 and 2 in Table 3 present relevant results from the event-history analyses of adult mortality.⁸ In Model 1 without fixed effects, coresidence with a grandfather or a grandmother alive at birth does not have a significant effect. However, the patterns change after unobserved characteristics of the family are accounted for in Model 2 by the introduction of fixed effects. Consistent with H2a, men whose grandmother was still alive when they were born have odds of dying 25.1 % higher than brothers born after that grandmother had passed away (e^0.224 = 1.251). In other words, among brothers who differed according to whether their paternal grandmother was still alive when they were born, those whose grandmothers were still alive had higher death rates in adulthood. An association between adult female mortality and community characteristics may have been suppressing the effect of coresidence of grandmother in Model 1. Meanwhile, no significant effects are found for grandfathers after the introduction of father fixed effects. For old age, we did not find any significant effect of coresidence with grandfathers or grandmothers at the time of birth on mortality.

Coresidence With Grandparents Later in Childhood

We explore whether the immediate or long-term effect of coresidence with grandparents in childhood varies by the duration of coresidence. The results are shown as Models 3 and 4 in Tables 2, 3, and 4, for childhood, adulthood, and old age, respectively. When we break down the effect by the age of the child, we find an adverse effect of coresidence with grandfathers and grandmothers for children who are not yet 5. However, the magnitudes of these effects become smaller after the introduction of father fixed effects, and they are no longer statistically significant. Comparison with the results for Model 2 in the same table confirms that coresidence with grandparents has no effect in childhood after comparison is made between siblings who differed according to age at the loss of their grandfather.

Coresidence with a paternal grandmother later in childhood has long-term adverse effects on adult mortality (H2a). According to Model 4 in Table 3, when a fixed effect of father is included, coresidence with a grandmother has adverse effects on adult mortality regardless of the duration of coresidence. The odds of dying are 22 % (e^0.199 = 1.220), 26 % (e^0.231 = 1.260), and 34.2 % (e^0.294 = 1.342) higher than brothers whose grandmothers were dead when they were born when coresiding with a grandmother at 1–5 sui, 6–10 sui, and 11 sui and older, respectively. These results indicate that the longer a man coresides with his grandmother in childhood, the higher the mortality risk he has in adulthood. This result reflects comparisons between brothers and is net of intergenerational correlations in longevity. Results for the effects of coresidence with grandfather are inconclusive. According to the results in Model 3 of Table 3, men whose grandfather was last seen alive when they were aged 11 sui or older have a lower mortality risk. In Model 4, which includes father fixed effects, this beneficial effect disappears. Long-term effects of coresidence with grandparents in childhood are not apparent in old age, as shown in Table 4.

We present the results of robustness checks for adult mortality in Table S5, Online Resource 1. We first include household context variables indicating the presence of father and mother around the time of the individual’s birth. We also experiment with inclusion of variables indicating parents’ longevity, measured by whether a father or a mother lived longer than age 60, to capture cases where parents lived to exceptionally advanced ages. The results are consistent with those in Table 3.

Interaction Between Loss of Parent Around Time of Birth and Presence of Grandparents

For men who experienced loss of their mother around time of birth, coresidence with a grandmother had a strong beneficial effect on adult mortality. By contrast, coresidence with both a grandmother and a mother at the time of birth has an adverse effect on adult mortality. These patterns are consistent with the in-law conflicts hypothesis (H2a). As shown in the fixed-effects model in Table 5, for an adult aged 16–55 sui, compared with brothers whose mother was alive around their time of birth but whose grandmother was no longer alive, men who lost their mother around their time of birth but had a coresiding paternal grandmother had 37.8 % (1 – e^{(–0.941 + 0.232 + 0.234} = 0.622, p = 0.053) lower odds of dying in the next three years. Adult males whose mothers and grandmothers were both alive at the time of birth have a 26.4 % (e^0.234 = 1.264) higher odds of dying in the next three years than men whose mother was alive but not their grandmother. The model includes controls for birth order, so these results are not an artifact of the fact that boys who lost their mother are by definition last-born. No significant results are found for coresidence with grandfathers.

The results suggest that despite an overall adverse effect of coresidence with a paternal grandmother, in the specific case where a mother passed away right after a boy was born, the presence of a paternal grandmother was beneficial. Either the grandmother substituted for the mother in providing care for the child, or she was able to advocate for the child in household negotiations over the allocation of resources. Moreover, the finding that mortality was highest for men whose mother and grandmother were both alive around the time of their birth, and was lower if only their mother or only their grandmother was alive, is consistent with the suggestion in H2a that elevated mortality associated with the presence of a grandmother was at least partly due to conflict between the mother and her mother-in-law.

Conclusions

Childhood coresidence has consequences for mortality at later ages. Coresidence with a paternal grandmother in childhood raises mortality risks in adulthood (H2a), and this adverse effect increases with the duration of coresidence. All these findings are robust to the inclusion of controls for family and household characteristics in childhood and adulthood, including SES and early household context as well as common effects of unobserved characteristics within a household. These results are consistent with the scarring channel proposed earlier (H2a). Results were not consistent with a selection process (H3), beneficial effects of presence of a paternal grandfather associated with their labor contributions (H1b), or beneficial effects of a paternal grandmother associated with assistance they provided to mothers (H2b). Although coresidence with a grandmother in childhood has an adverse effect on mortality in adulthood, boys who lived with either their mother or grandmother but not both around the time of birth had lower mortality in adulthood, which is consistent with the suggestion in H2a that conflict between a mother and her mother-in-law could produce harmful effects. No significant effects on mortality are found for coresidence with paternal grandfathers in childhood. Effects of coresidence with paternal grandmothers on later-life mortality vary according to the duration of coresidence, consistent with the idea that there were critical periods.

This study contributes to the literature on the long-term consequences of early-life context in several ways. It considers a previously understudied topic: the long-term influence of coresidence with grandparents early in life. Although grandparent effects on child well-being are already the subject of a very large literature, and grandparent effects on socioeconomic attainment are the subject of a rapidly growing literature, few studies have explored the effects of childhood coresidence with grandparents on health in adulthood and old age.

Our study is also distinctive in that it uses data from a preindustrial, primarily rural society, where mortality was high and the household was the most important unit of social organization. Although the data have limitations that affect our ability to generalize—most notably because of the inability to follow women across the life course and measure the effects of coresidence in childhood on their elderly mortality—the results nevertheless demonstrate the potential importance of a mechanism that should be studied in other settings where better data may become available.

An important implication of these results is that in extended families in patriarchal societies, patterns of coresidence in childhood may have long-term consequences for health and mortality, net of common environmental and genetic factors. Significance of the effect of coresidence with grandmothers after introduction of fixed effects tells us the association found is not a product of unobserved common household features. A more general implication of these results is that they confirm that studies measuring grandparent effects on socioeconomic and health outcomes later in life need to consider coresidence in childhood or other exposure as an intervening variable (Song and Mare 2017), at least in societies where multigenerational households are common.

Our findings may also have special implications for understanding health and mortality among the elderly in the many non-Western societies where coresidence with grandparents was common when today’s elderly were still young. Although societies characterized by patrilocal residence and an ideal of multigenerational coresidence such as India and China are the most obvious examples, some form of coresidence with grandparents was also common elsewhere in Asia and Africa. The social context under our examination is a preindustrial rural society where mortality was high and the household was the most important unit of social organization, which is comparable to the environment in which the elderly in many contemporary developing countries spent their childhood. More generally, our results suggest that contemporary mortality among the elderly in developing countries may reflect the legacy of coresidence in childhood many decades ago. To the extent that mortality is reflective of underlying health, the implication is that contemporary differences in health among the elderly may reflect differences in their experience of coresidence in childhood. Obviously, the social and cultural context of coresidence in other settings would differ, and we hope that as more data become available, comparison between settings will help illuminate specific mechanisms driving these associations.

Acknowledgments

We are grateful to Dwight Davis, Hao Dong, Noreen Goldman, James Lee, Evan Roberts, Xi Song, and members of the Lee-Campbell research group for their suggestions. Versions of this article were presented at the annual meeting of the Population Association of America, Boston, MA, May 1–2, 2014; and the Social Science History Association Annual Meeting, Toronto, ON, November 6–9, 2014. Preparation and documentation of the China Multi-Generational Panel Dataset, Liaoning (CMGPD-LN) for public release via ICPSR Data Sharing for Demographic Research (DSDR) was supported by NICHD R01 HD057175-01A1 “Multi-Generation Family and Life History Panel Dataset” with funds from the American Recovery and Reinvestment Act.

Notes

References