Abstract
In addition to own education and other socioeconomic resources, the education of one’s children may be important for individual health and longevity. Mothers and fathers born between 1932 and 1941 were analyzed by linking them to their children in the Swedish Multi-generation Register, which covers the total population. Controlling for parents’ education, social class, and income attenuates but does not remove the association between children’s education and parents’ mortality risk. Shared but unmeasured familial background characteristics were addressed by comparing siblings in the parental generation. In these fixed-effects analyses, comparing parents whose children had tertiary education with parents whose children completed only compulsory schooling (the reference group) yields a hazard ratio of 0.79 (95 % CI: 0.70–0.89) when the socioeconomic position of both parents is controlled for. The relationship is certainly not purely causal, but part of it could be if, for example, well-educated adult children use their resources to find the best available health care for their aging parents. I therefore introduce the concept of “social foreground” and suggest that children’s socioeconomic resources may be an important factor in trying to further understand social inequalities in health.
Introduction
People’s own socioeconomic resources (Elo 2009; Geyer et al. 2006; Torssander and Erikson 2010), those of their partners (Lager et al. 2011; Skalická and Kunst 2008; Torssander and Erikson 2009), and those of their parents (Galobardes et al. 2004; Kuh et al. 2002; Næss et al. 2004) have all been shown to be related to health and length of life. However, research into the relationship between children’s socioeconomic resources and their parents’ longevity is uncommon. The few studies that have been carried out suggest that children’s education is associated with parents’ mortality risk in the United States (Friedman and Mare 2010) and in Taiwan (Zimmer et al. 2007). The present study explores this relationship further.
An association between children’s education and their parents’ health is, of course, to be expected because children’s level of education largely reflects the socioeconomic resources in the parental household. Hence, studies of the importance of children’s resources for parents’ health need to extensively address these confounding factors, which can be difficult. Moreover, adult children today are seldom part of the household of their aging parents, which suggests that the socioeconomic resources of partners are more important. However, it may nonetheless be reasonable to assume that adult children’s resources—the “social foreground”—also matter for parental health and mortality.
The potential importance of children’s resources for parents’ health is most likely greater in countries where coresidence of adult children and their parents is common. However, even though adult children and their parents generally live separately in Sweden and other Western countries, four of five adult Swedes talk with their parents every week, and three of four live within 100 km of their parents (Lennartsson et al. 2010). Thus, adult children constitute a significant aspect of their parents’ social network, and the structural conditions facilitate transmissions of resources between the generations.
Theories about social networks and health suggest various pathways through which social contacts may affect health (Berkman and Glass 2000). Three of these pathways are potentially relevant for child-to-parent transmission: provision of social support, social influence, and access to resources.
The first pathway, social support, includes various types of support, such as emotional, instrumental, and informational support. Those in working-class families seem to have more contact across the generations than those in upper-middle-class families, at least in Sweden (Fors and Lennartsson 2008). The frequency of contact may be of importance, for example, if parents have minor health problems and are in need of regular care and help with practical issues: that is, instrumental support. In addition, increasingly since the 1980s, older Swedes with nonsevere health conditions rely partially on help from family members or the private market, with a substantial amount of informal care and support given to older parents by children or partners (Szebehely 1998). If highly educated children have less time for their parents, children’s education may not be positively related to this practical kind of social support. In the case of daughters, for example, women with a low level of education are more likely to work part-time than more highly educated women (Evertsson et al. 2009), even though the latter group generally has more flexible working conditions. Daughters are more involved in giving time-consuming help than sons (Fors and Lennartsson 2008), and relationships with mothers seem to be more cohesive than relationships with fathers (Silverstein et al. 1997). Thus, gender of both the child and the parent seems to play a central role here.
The characteristics of the informational support given to parents may differ according to the educational level of their children. Informational support refers, for example, to the provision of advice or guidance by close others. Although the frequency of contact may sometimes be important, at other times, the ability of well-educated adult children to give health-related advice and deal with the health care services may be of greater value—for example, if a parent is more seriously ill. Hence, parents with well-educated children may have an advantage where some types of social support are concerned, even if this is not the case with frequent practical help.
The second pathway, social influence, is another potential causal mechanism between children’s education and parents’ health and longevity. For example, well-educated individuals are less likely to smoke than people with a lower level of education (de Walque 2007). Thus, children who are better educated than their parents may adopt health-related behaviors that differ from those of their family of origin. In turn, children’s newly adopted health behaviors may influence parental habits, for example, regarding smoking, but also for other health-related behaviors, such as diet, exercise, and drinking. Research into this kind of spillover effect from adult children to parents is scarce, but a related study has suggested that the partner’s education is negatively associated with smoking (Monden et al. 2003). Although such an association may be partly due to assortative mating (i.e., choice of partner may be influenced by the partner’s health behavior), it is also possible that close family members influence each other to give up smoking or other unhealthy habits and that there could thus be shared family norms regarding health behaviors. In addition, U.S. data have shown that the relationship between children’s education and parents’ survival can partly be explained by parents’ health behavior (Friedman and Mare 2010). This correlation does not necessarily represent causality, but if it to some extent does, parents with well-educated children may have an advantage.
The third pathway, access to resources, traditionally refers to material resources. Strictly financial upward intergenerational support is, however, uncommon in Sweden (Fritzell and Lennartsson 2005). Although economic resources may be a more straightforward kind of transfer between family members than education, it may nevertheless be possible to share the nonmaterial returns of education. At the individual level, it has been suggested that education increases individuals’ understanding of health issues, or at least that schooling improves skills that enable them to apply messages about health to their own lifestyles (Cutler and Lleras-Muney 2010). When adult children provide their parents with various forms of support, it is likely that children’s knowledge or related skills are put to use. They may also help their parents with health-related advice when specific knowledge or the ability to assimilate information is important. Examples of this might be health behaviors (eating, exercising, smoking, and drinking) and when and how to make good use of available health care. It has also been suggested that well-educated children may be better informed about the latest health technologies than their less-educated counterparts, and indeed better informed than highly educated parents themselves (Friedman and Mare 2010).
Previous studies on social inequalities in health care utilization show varying results (summarized in Burström 2002). However, it has more recently been shown that people with higher incomes are more likely to see a specialist than people with lower incomes (van Doorslaer et al. 2006), and well-educated patients receive newer and more expensive prescription drugs than patients with a lower educational level (Wang et al. 2007; Weitoft et al. 2008). Such differences between individuals in the utilization of health care may also differ according to the educational level of adult children. For example, adult children may, as mentioned, support ill parents in their contacts with health care professionals and push for the fastest and best available care, and perhaps children with more educational resources have greater potential to ensure that the available medical care is fully used. Thus, education can be viewed as a family resource, with parents benefiting from adult children’s education-related abilities.
Besides the aforementioned pathways, there may well be other mechanisms. For example, perhaps parental worry about their adult children is an explanatory factor (Hay et al. 2008). If children are faring well—they have a job and are earning their living—parents may have less worries about them. And because well-educated children, on average, enjoy more stable circumstances than less-educated children, their parents may have less reason to feel concern. These parents may also be able to retain more of their own (economic) resources.
It has also been suggested that status feelings (as a general term for social standing in society related to prestige and esteem) may influence health and, in the long run, also longevity (Marmot 2004). If status in this general understanding is shared within the family—say, through having a successful spouse—and is not solely based on own achievements, successful adult children may similarly raise their parents’ subjective social status. In this sense, the parents of successful children may acquire status through their offspring.
Causality, Reverse Causality, or Confounding Factors?
All these suggested mechanisms presume a causal link between children’s education and parents’ health and survival—for example, through health-related advice, social influence, and/or support in navigating health services. Research into the association between own education and health sometimes indicates causality; some (although not all) studies using schooling reforms as exogenous variation in the length of compulsory education show that more schooling leads to better health (Eide and Showalter 2011). Another explanation may be reverse causality (Case et al. 2005; Currie 2009)—that is, if poor health prevents a person from continuing to higher education. Finally, there may be common causes that determine both education and health; such confounding factors include cognitive ability, social background, and personality characteristics.
All these explanations are applicable, at least to some extent, when it comes to the role of children’s education for their parents’ longevity. If parental health affects not only one’s own education but also that of the next generation, reverse causality may be the explanation. For example, growing up with ill parent(s) could hamper the child’s schooling because the parent may not be able to give the help that a healthy parent would, and the child may be obliged to take more responsibility in the home. The household’s economic circumstances may also be more strained, which would in turn affect the child’s opportunity to continue on to higher education. This may be of less importance in Sweden, however, where tertiary education is free of charge.
In addition, many factors may influence both children’s education and parental health. The parent’s own socioeconomic resources are of greatest importance, but the resources of the parent’s partner also play a role, especially when the partner is the other parent of the child. Other factors that may confound an association between children’s education and parental health include parents’ cognitive ability, parental social background, and parents’ preferences for long-term investments. Such preferences could confound the relationship if some parents choose to invest both in their children’s education and in their own health. Preferences for long-term investments have been put forward as an explanation for the relationship between own education and health (Fuchs 1982), albeit with limited empirical support (Cutler and Lleras-Muney 2010).
Finally, there may be a causal effect if children’s education actually affects their parents’ health through any of the pathways suggested earlier—for example, receiving help and support in health-related situations.
The hypothesis tested in the present study is that children’s education is negatively associated with parental mortality risk. Moreover, this association should hold net of the parent’s own education and other socioeconomic resources. The present study controls for a range of the parent’s, and the parent’s partner’s, socioeconomic resources and is thus able to take the most apparent confounding factors into account. In addition, the interaction between children’s education and parent’s sex is examined to establish whether mothers gain more from their children’s education than fathers do. Previous research suggests, as mentioned earlier, that the child-mother relationship is stronger than the child-father relationship. Similarly, the gender of the child is taken into consideration because earlier studies suggest that daughters help their parents more than sons do.
It will be hard to prove that an observed association is causal even if it remains after the most important confounding factors are controlled for. However, it can at least be approached by means of a sibling-comparison design (Lawlor and Mishra 2009). This design enables a comparison of siblings in the parental generation by adding family-fixed effects, and reduces confounding from unobserved family characteristics. The sibling-comparison design is thus a way of adjusting for parents’ socioeconomic conditions during childhood, given that no such direct information is available for the parent generation. Although, for example, health behavior is ultimately an individual action, there could also be social constraints. Parents with similar early experiences and the same family background may share some of these constraints, and siblings may therefore form a better comparison group (i.e., have more in common) than unrelated individuals. Hence, the results from the sibling models will to some extent help to distinguish between what “runs in the family” and what is actually an effect of children’s educational level.
Method
Data
The study is based on individuals in the Multi-generation Register (Statistics Sweden 2008), which includes all Swedes born in 1932 or later and living in Sweden any year after 1960, and connects parents to their offspring. The sibling-comparison design—that is, comparing siblings in the parental generation—is therefore possible only for individuals born in 1932 or later. The main analytic sample consists of parents born in 1932 and onwards, with the upper limit set at 1941, because there must be time for their children to complete their education as well as a reasonable follow-up period in which to observe parental deaths. Parents who gave birth to their first child comparatively late had to be excluded because their children would not have completed their highest education sufficiently long before the end of the parental mortality follow-up period. Consequently, the youngest child cohort included was born in 1969, which excludes 10.1 % of the parents born between 1932 and 1941. Two other criteria for inclusion in this main analytic sample is that they were alive and living in Sweden in 1970, the census year for which information on socioeconomic factors was collected. Finally, the registers must hold information about the parents’ as well as their children’s educational attainment (this information is missing for 6.2 % of these parents).
Because of these necessary restrictions, the main analytic sample consists of 624,761 parents (290,023 fathers and 334,738 mothers) who had their first child when they were relatively young and for whom there is register information on both their own and their firstborn child’s education. Among these parents, it is possible to identify 164,553 full biological siblings with the information in the Multi-generation Register.
Information about own education and occupation was obtained from the census of 1970, when the parents were aged 29–38. For individuals who were not gainfully employed at that time, occupational information from the censuses of 1975 or 1980 was used, when the parents were aged 34−48. This strategy was used because many women did not work when their children were small but returned to the labor market later on. In addition, the partner’s occupational class is included in the analyses, giving a more complete picture of the position of the family.
Data on income from work were taken from the Swedish Income and Taxation Register for the 10 years between 1970 and 1979, giving a more stable income measure than annual income. Annual income was recalculated according to the consumer price index in 1979 to take inflation into account.
Cohabiting and/or married partners—when this partner is the other biological parent of the child, according to the Multi-generation Register—were identified in the 1970 census for 83.1 % of the main sample (248,078 fathers and 270,974 mothers), and socioeconomic information for partners was collected from the same year. Information about the educational level of the children was derived from educational registers between 1985 and 1999, depending on the child’s birth cohort (the year closest to the one in which the child reached age 30).
The parents were followed up in the Cause of Death Register until the end of 2007, when the oldest individuals were 75 years old (in total, 84,635 deaths: 13.5 % of 624,761 parents). The person-unique Swedish identification number served as the key match variable between the registers used.
In addition, old-age mortality risks for some cohorts of parents were analyzed, although it was not possible to adjust for sibling fixed effects here because of the limits of the Multi-generation Register. These parents were born between 1915 and 1931 (and their children were born between 1932 and 1969) and composed a sample size of 552,158 fathers and 629,510 mothers. In these earlier parental generations, 99.4 % had their first child before 1970; therefore, there is (in this case) no problem with selection of individuals who became parents at relatively young ages.
Variables
Education refers to the highest attained level of education according to the following categories: (1) compulsory schooling; (2) shorter upper-secondary schooling (less than three years, usually vocational training); (3) longer upper-secondary schooling (three years or more); (4) tertiary education of less than three years; and (5) tertiary education of three years or more. Education was measured for children, parents, and partners.
Because most parents have more than one child, it is not obvious how children’s education should be summarized into one measure. As mentioned earlier, there is a selection of firstborns because there must be time both for children to complete their education and also for a follow-up period for parental deaths. Because of this selection, I have chosen to use the education of the firstborn child of each parent. One could, alternatively, adopt a dominance approach: that is, using the highest attained education of all the parent’s children—or the proportion of, for example, children with tertiary education. These strategies are also tested.
Parental social class is based on a grouping of occupational codes (NYK, Nordic standard occupational classification), available from the censuses, into a Swedish socioeconomic classification (SEI). This classification resembles the EGP (Erikson-Goldthorpe-Portocarero) class scheme (Erikson and Goldthorpe 1992) and distinguishes seven groups: unskilled manual occupations, skilled manual occupations, lower nonmanual occupations, intermediate nonmanual occupations, professional and managerial occupations, farmers, and other self-employed (including self-employed professionals). Social class is used here as an individual measure, with those who were not gainfully employed in any of the censuses coded as missing. However, because social class of partners is included, the combined social class of the household will be captured.
Parental income is measured as average individual income from work during the years 1970–1979 and includes tax-related benefits. In the analyses, income is divided into quintiles for men and women separately.
In addition, number of children is included as a control variable in the analyses because the number of children in a family has been shown to be associated with both the main independent variable, children’s education (Downey 1995), and parents’ mortality risk, albeit with inconclusive results (Grundy and Kravdal 2008; Hurt et al. 2006; Tamakoshi et al. 2010).
Statistical Analyses
Cox regressions were used to estimate hazard ratios of mortality for the different groups. Parental age (in months) is the time variable, and parental death is the event. Censoring occurs at emigration, child’s death, or December 2007. Parents enter the analyses when their firstborn child is 31 years of age because highest attained education is measured when the child is 30. Parents who die or emigrate before their child reaches 31 years of age, as well as those whose children die before age 31, were excluded from the analyses. Fixed effects at the family level, with all full siblings in the parent generation being given the same family ID, were added in the most restrictive models. These results can be interpreted as a comparison between siblings whose children have different levels of highest attained education.
Results
Descriptive statistics for the independent variables are shown in Tables 1 and 2. The compositional change in education over time is reflected in the proportions of parents (Table 1) and children (Table 2) in each educational group. The numbers of those with only compulsory education decreased noticeably, from 65 % (mothers) and 52 % (fathers) to about 16 % in the child generation. All groups with higher education are larger in the child generation than in the parent generation. Moreover, the gender difference in educational attainment found in the parent generation—with fathers, on average, being better educated than mothers—has almost disappeared in the child generation, where it is actually slightly more common for men to have achieved no more than compulsory education.
In the sibling comparison model, full sibling groups within this sample are analyzed (N = 164,553). Descriptive statistics for this more restrictive sample are shown in Table S1 of Online Resource 1. Overall, the distributions of the variables are similar to those in the full sample, indicating that these individuals do not differ substantially from the population as a whole. If anything, individuals in the sibling sample are less educated than the total population: 56 % of the male siblings identified have compulsory education only, compared with 52 % in the full sample; the corresponding figures for women are 67 % and 65 %, respectively. This difference between the sibling sample and the full sample may be due to the negative association between number of siblings and educational attainment (Downey 1995) and to the increased probability of inclusion in the sibling sample associated with having many siblings. However, there are no appreciable differences between the sibling sample and the full sample in the distribution of the independent variables.
Without controlling for the socioeconomic circumstances of the parents, there is a clear and gradually decreasing risk of dying as the level of schooling of the children rises (Model 1, Tables 3 and 4). For both mothers and fathers, having at least one child with a minimum of three years of tertiary education yields a hazard ratio (HR) of 0.56 (mothers, 95 % confidence interval (CI): 0.54 to 0.58) and 0.58 (fathers, 95 % CI: 0.56 to 0.60) compared with having a child who completed only compulsory education (the reference category). Model 1 includes a control for how many children the parent has, with results indicating that having only one child is associated with an increased risk of dying (HR = 1.30 for mothers and HR = 1.23 for fathers) compared with the reference category of two children. Having four or more children is also associated with an increased mortality risk for both mothers and fathers, demonstrating the U-shaped relationship between number of children and mortality risk found elsewhere (Tamakoshi et al. 2010). However, controlling for number of children does not essentially change the bivariate association between children’s education and parents’ mortality risk: the corresponding hazard ratio for mothers and fathers of children with compulsory education is, respectively, 0.55 and 0.56, without controlling for number of children (not in Table).
The results from Model 1 will reflect, and be influenced by, the parent’s own education and other socioeconomic resources in the household. Model 2 includes parental education, with slightly reduced estimates of children’s educational level: the hazard ratio is 0.61 for mothers and 0.66 for fathers in the most-educated group. Notably, the death risk gradient is as steep as (fathers) or steeper (mothers) across children’s education compared with own education. This should not, however, be interpreted as indicating that children’s education is equally or more important than own education. Rather, children’s education is largely a function of socioeconomic resources in the household: that is, not only of one parent’s education. Moreover, there are large compositional differences between generations, with the reference group (compulsory schooling) being much smaller for the child generation than for the parent generation; it is therefore possible that selection into this group differs between generations.
Adding parental occupational class and average income from work during the period 1970–1979 (Model 3) attenuates the estimate for the child’s education for fathers (from 0.66 to 0.71) but not for mothers (0.61 to 0.61). Two factors may explain this. First, father’s class and income are more closely related to children’s educational attainment than mother’s class and income are. Second, father’s class and income are better proxies of the household’s overall resources than mother’s class and income are. The finding in Model 3 that mother’s own education, but generally not her income, is related to her death risk may be due to problems of multicollinearity or to the fact that these factors have no independent relationship with women’s mortality when education is included; it is also consistent with prior results based on a much larger sample for which multicollinearity is less of a problem (Torssander and Erikson 2010). However, to return to a possible child effect on parental mortality, much of the original association between children’s education and parental death risk remains in Model 3 and cannot be accounted for by any of the parental socioeconomic measures. In sum, the excess mortality risk for mothers of children who did not continue their education after compulsory school falls when the parent’s own socioeconomic resources are added to the model (from 0.56 to 0.61), and this reduction is more pronounced for fathers (from 0.58 to 0.71).
The next step was to add measures of socioeconomic position of partners (i.e., the other parent of the child), since these factors are related to children’s education as well as parental longevity. This further attenuated the association between children’s education and parental mortality risk. However, much of the association between children’s education and parents’ longevity remains when the socioeconomic position of the other parent has been adjusted for. In this model (Model 4, Tables 3 and 4), parents of children with tertiary education have an approximately 30 % lower hazard than parents of children who did not continue after compulsory school. In addition, significant mortality differences still exist across all educational groups of children compared with the reference category. The association between children’s education and parental death risk is somewhat stronger for mothers than for fathers in Model 4 (HR = 0.67 vs. 0.74), which might give support to the hypothesis that mothers gain more from having well-educated children than fathers do. This is to be expected if the mother-adult child relationship is closer than that between father and adult child, and if adult children help their aging mothers more than their aging fathers. However, pooling mothers and fathers and including an interaction term between children’s education and parent’s sex gives a nonsignificant result. It is nevertheless possible that mothers living alone in old age (not possible to study here) gain more from their children’s education than single aging fathers do.
Comparing Siblings
The multivariate Models 3 and 4 were run again for identified full biological siblings within this sample of parents born between 1932 and 1941 (with at least one child born before 1969). Siblings in the parent generation seem to have about the same death risks across educational groups as parents without siblings with children in these cohorts.
In the sibling models (Table 5), mothers and fathers are pooled because some of them are siblings. (A dummy variable of parent’s sex is, however, included.) Adding sibling fixed effects reduces mortality differences between educational groups; the hazard ratio for parents with children with longer tertiary education becomes 0.74 (compared with 0.69 for this sample without including sibling fixed effects). Hence, the association between children’s education and parents’ death risk is partly explained by factors that these siblings share, although they do not play a major explanatory role. In the most comprehensive model (Model 3, Table 5), partner characteristics are also controlled for. Comparing siblings with not only the same socioeconomic characteristics but also comparable partner characteristics reduces the association between children’s education and parents’ mortality further (HR for longer tertiary education = 0.79; 95 % CI: 0.70 to 0.89). Hence, there is still an unexplained difference in mortality risk of about 20 % between the lowest and the highest educational group.
In conclusion, the relationship between children’s education and parents’ mortality can be only partly explained by the parent’s or partner’s/other parent’s socioeconomic position (measured as education, social class, and income). Moreover, the relationship can be explained only to some extent by factors that siblings share with each other, such as family background.
Further Analyses
Alternative Measures of Children’s Education
The analyses so far show only the educational level of the oldest child in the family. It is unclear whether it is the oldest child who gives the most support to the parents or the extent to which other siblings help out as well. Using a dominance approach when deciding the education of children or a measure of a share of children with tertiary education produces similar results (Table S2, Online Resource 1). However, looking at parents with only one child may be a better robustness test because their education is a more exact measure of the resources in the social foreground. The hazard ratios from the analyses of parents with one child (Table S3, Online Resource 1) are similar to those based on all parents independent of how many children they have. Thus, using the education of the oldest child as a proxy for the educational level of all children seems, in this respect, to be an acceptable alternative.
Gender of the Child
For the parents with one child only (N = 96,550), having a daughter is related to a slightly lower death risk than having a son (HR = 0.95; 95 % CI: 0.92 to 0.98). Hence, the sex of the only child seems to influence the mortality risk, possibly because daughters may be more likely than sons to help their ill and aging parents. The pattern is not as clear for parents with several children, where the hazard ratios are nonsignificant. However, there is no significant interaction between the gender of the child and the child’s educational attainment. Thus, according to these results, daughters’ education is not more important for parental health/mortality than that of sons, perhaps because highly educated women tend to work more hours than less-educated women.
Earlier Parental Generations (Born Between 1915 and 1931)
Because the hypothesis regarding differences in health care utilization according to children’s education primarily concerns a somewhat older population that tends to use more health care, analyses were also carried out on 17 birth cohorts of parents who were aged 76 to 92 at the end of the follow-up period. Of these older parents, 54 % died during the follow-up period: 87 % of the 1915 cohort, and 24 % of the 1931 cohort (numbers are conditional on survival until 1970, when the independent variables were measured). Compared with the 1932–1941 cohorts, there are very few problems with younger parents being overrepresented: only 0.6 % of these parents’ firstborn children were born after 1969 (with the consequence that the children would have been too young to have completed their highest education before the start of the follow-up). Cox regressions based on these older cohorts demonstrate that when own education, social class, and income in 1970 (for the partner, also) are taken into account, there is still an association between children’s education and parents’ relative death risk: among parents in the highest educational group, the hazard ratio for fathers is 0.82 (95 % CI: 0.81 to 0.83), and the hazard ratio for mothers is 0.78 (95 % CI: 0.77 to 0.80; see Model 2, Table S4, Online Resource 1). As mentioned earlier, it was not possible to add family fixed effects and compare siblings in the parent generation because the Multi-generation Register includes only parents of individuals born after 1932. The educational attainment among these older parents and their children diverges from that of the younger cohorts. As many as 71 % of these parents had no more than compulsory education, and the corresponding figure for their children is 28 %.
Discussion
Before the introduction of the welfare state, children did, of course, play a crucial role in helping their ill and aging parents. However, according to empirical research, even after the creation of the public systems offering economic security and health care to the elderly, upward intergenerational contact and support between adult generations remained (Fors and Lennartsson 2008; Silverstein et al. 1997). “The decline of the family” is nevertheless true in the sense that intergenerational living arrangements that include grandparents are rare in Western societies today. On the other hand, because of increasing longevity, it is not exceptional for three adult generations to be alive at the same time, and there has thus been an increase over time in the number of years during which adult children and their parents coexist.
The relationship between own schooling and health is well established, and family socioeconomic resources—in the shape of resources of partners and parents, for example—are also related to health and longevity. Nevertheless, with a few exceptions (Friedman and Mare 2010; Zimmer et al. 2007), researchers have often dismissed the resources of adult children. The present study demonstrates that the relationship between children’s education and their parents’ longevity cannot be fully explained by the socioeconomic resources of either the parents or their partners (where the partner is the other parent of the child). The hazard ratio in the most comprehensive model that includes family fixed effects for the parent generation is 0.79 (95 % CI: 0.70 to 0.89) for a parent with a child with longer tertiary education compared with a parent of a child with compulsory schooling only. Thus, even when considering the socioeconomic resources of parent and partner and comparing parents who are full siblings, an association between children’s education and parents’ death risk remains, corresponding to a 21 % lower hazard during the follow-up period for the group of parents with the most highly educated children.
Nevertheless, the underlying reasons for this association may not be causal because it is impossible, in practice, to adjust for all potential confounders in observational studies. However, full siblings might serve quite well as controls for one another: they share some genes, and in most cases, they grew up in the same family and have consequently experienced a similar environment and socioeconomic conditions during childhood and adolescence. A family-based study design could therefore be more fruitful than one based on unrelated individuals. However, there are, of course, also differences between siblings that cannot be detected, which open up the possibility of noncausal explanations.
In addition, the present findings should be viewed in relation to the Swedish context. Sweden introduced a comprehensive elementary school system with fairly late tracking in the middle of the twentieth century (Jonsson and Mills 1993). There are no tuition fees for tertiary education, and grants and loans are available to cover living costs. Other national institutional characteristics may also be of importance when interpreting the robust relationship between children’s education and parents’ mortality found here. For instance, when social policies such as those related to health care and pensions are comprehensive, the resources of other family members may be of less significance among the elderly than when social policies are more restricted and targeted. The Nordic countries, including Sweden, have developed welfare states characterized by universal social policies and “equality of the highest standards” (Esping-Andersen 1990:27). Equality of access to health care with a maximum fee charge since the 1970s (Andersen et al. 2001) and an all-encompassing pension model based on a universal basic pension and previous earnings (Korpi and Palme 1998) are two examples of institutional characteristics that may curb the importance of family members’ support in old age. But, as mentioned previously, some research indicates that in practice, health care is not equal in Sweden. For example, the better off a person is, the more likely he or she is to see a specialist (van Doorslaer et al. 2006). Nevertheless, the comparatively egalitarian system may make Sweden a conservative test case for the relationship between children’s education and parental mortality; that is, in countries with less universal social policies, family members’ resources may play a more important role for the elderly.
In studies of the relationships between social background, own socioeconomic position, and later life chances, the time order per se (i.e., childhood factors preceding later-life health) might suggest causality. This is not the case when one looks at the social foreground and its effects on health. One possible alternative explanation is, hence, reversed causality: that is, that the poor health of parents influences their children’s educational attainments—for example, because they receive less help with homework or are exposed to emotional stress. However, because there is no charge for higher education in Sweden, low income because of illness in the family does not have to be a major obstacle to university education for children in such circumstances. Furthermore, similar results are obtained when only mothers and fathers who were in the labor market and were accordingly well enough to be able to work are included. Thus, in Sweden, lack of money among ill parents is probably not a major confounding factor. Moreover, even parental deaths have only a small negative effect on children’s educational attainment, according to previous research on Swedish data (Adda et al. 2011). However, this design does not completely rule out reversed causality, which suggests a cautious approach when studying this relationship further. In addition, chronic health conditions during childhood could prevent the parent from obtaining a higher education (Case et al. 2005), which in turn may have an impact on their children’s educational attainment. This possibility should not be disregarded.
Although the most apparent confounding factors (education, occupational class, and income of both partners) were considered in the analyses, other omitted variables may influence the result. For example, children’s education may be a reflection of parental abilities not manifested in the parents’ own educational and occupational achievements. One explanation for the observed relationship between children’s education and parental mortality could therefore be that personal characteristics, such as cognitive resources or ambitions/aspirations, influence both factors. The association between children’s education and parents’ death risk would then be a mere reflection of nonutilized parental abilities. In the present study, I attempted to examine such abilities by comparing full siblings in the parent generation, given that some of these unmeasured personality characteristics may be more shared among siblings than among totally unrelated individuals. However, the family fixed-effects model also yielded clear significant differences in parental death risk by children’s education. Hence, the possibility of an upward intergenerational transmission of educationally linked resources from child to parent cannot be excluded, although it is not clear exactly what siblings share (or do not share). Another possible explanation is that parents who prioritize investing for the future encourage their children to pursue higher education and also invest more in their own health. However, if such parental characteristics are an important explanation of the relationship between children’s education and parents’ health, they would not primarily be captured by own socioeconomic position because this is controlled for.
The finding that mothers do not gain more from their children’s education than fathers do cautions against concluding that there may be a causal effect of children’s education on parents’ mortality risk. Because mother-child relationships have been shown to be closer than father-child relationships (Silverstein et al. 1997), one would perhaps expect greater mortality disparities by children’s educational attainment for mothers than for fathers. However, where health differences according to children’s educational level are concerned, the frequency of contact may not be the only important factor, and adult children may perhaps give their mothers and fathers similar amounts of help when they are seriously ill.
Of parents with only one child, the mortality risk among those with daughters is about 5 % lower than for those with sons. This, if anything, is most probably a causal effect because parents do not select the sex of their child (although parents with one child could also differ from other parents in various ways). One explanation of this finding that is in line with previous research is that parent–adult daughter contact is more cohesive and frequent than parent–adult son contact. However, because there was no interaction between the gender of the child and the child’s educational attainment, this indicates that two different processes are operating and that both of them may influence parental death risks. The first process is about social contacts. Having a supportive network and seeing friends and family often has positive consequences for health, which benefits parents with daughters more than parents with sons. Second, in certain situations—for example, when medical care is needed—specific types of help can be crucial for health. In such situations, the gender of the child is perhaps of less importance than the child’s socioeconomic resources.
The issue of why own education shows such a strong and graded relationship with various health outcomes has been extensively discussed (e.g., Kawachi et al. 2010; Lleras-Muney 2005). Here, however, I would rather speculate about why and how children’s education may be important to their parent’s health beyond parental education and related resources. First, it should be noted that the Swedish educational system changed greatly between the parental and child generations in the present study. Compulsory education for the majority of these parents was only seven or eight years, but their children typically had nine years of compulsory schooling. In addition, both upper-secondary and tertiary education expanded rapidly in the 1960s (Murray 1988), which then influenced the child generation more than the parent generation. These structural changes are important when interpreting the present findings. For example, in the children’s generation, we might expect those who attended only compulsory school to be a more negatively selected group than in the parents’ generation. We might, however, also expect those with tertiary education to be a less-positively selected group of individuals among the children than the parents. Still, the implications of different educational levels are probably not identical for the two generations.
Moreover, the timing of education may impact certain types of knowledge—for example, in relation to technological change and scientific innovations, which relates to the health field. Here, a higher education attained comparatively recently may confer more up-to-date knowledge. In addition, education may in some sense lose its meaning for an ill and old individual, for whom the ability to self-advocate and actively use one’s resources may be reduced. In such situations, resources of other close family members may be of more manifest importance. Although knowledge-related resources such as education are traditionally viewed as personal resources, they could potentially be pooled within the family, as earlier suggested for partners (Lager et al. 2011). The present results indicate that this may be the case, but more research is needed to explain why and how these mechanisms operate. In addition, the resources of adult children may be of greater importance in countries with a less-comprehensive welfare system than Sweden. This is also a question for future research. With growing numbers and proportions of older people in the population, and given that a per capita reduction in spending on welfare is plausible, adult children’s resources will perhaps be even more crucial for parental health and longevity in the future. Although most elderly Swedes are not economically vulnerable (Fritzell and Lennartsson 2005), adult children may, in some situations, provide resources that the parents lack or perhaps add to the resources that the parents already have. For an ill and aging parent, it may be important to have a child with good resources who can help in a variety of health-related situations. The empirical results of the present study suggest that not only our social background but also our social foreground may contribute to an understanding of inequalities in health and longevity.
Acknowledgements
I thank Robert Erikson, Viveca Östberg, Juho Härkönen, Are Skeie Hermansen, Laust Hvas Mortensen, Denny Vågerö, three anonymous Demography reviewers, and participants at seminars at Stockholm University and the RC28 conference in Essex 2011. This study was financially supported by the Swedish Council for Working Life and Social Research (FAS, nr 2010-0101).
