Contemporary stratification research on developed societies usually views the intergenerational transmission of educational advantage as a one-way effect from parent to child. However, parents’ investment in their offspring’s schooling may yield significant returns for parents themselves in later life. For instance, well-educated offspring have greater knowledge of health and technology to share with their parents and more financial means to provide for them than do their less-educated counterparts. We use data from the 1992–2006 Health and Retirement Study (HRS) to examine whether adult offspring’s educational attainments are associated with parents’ survival in the United States. We show that adult offspring’s educational attainments have independent effects on their parents’ mortality, even after controlling for parents’ own socioeconomic resources. This relationship is more pronounced for deaths that are linked to behavioral factors: most notably, chronic lower respiratory disease and lung cancer. Furthermore, at least part of the association between offspring’s schooling and parents’ survival may be explained by parents’ health behaviors, including smoking and physical activity. These findings suggest that one way to influence the health of the elderly is through their offspring. To harness the full value of schooling for health, then, a family and multigenerational perspective is needed.
The association between one’s own socioeconomic status (SES) and health and longevity is well documented (Liang et al. 2000; Lynch 2003; Mare 1990; Smith and Kington 1997). Higher levels of schooling, in particular, are associated with better health, even more so than income. Despite an extensive body of work on the relationship between SES and health, however, few studies have considered the SES of the family—that is, beyond that of the married couple,—and its relationship with health and mortality. This article investigates whether the educational attainments of individuals should be viewed as a family resource, benefiting not only the individuals themselves but also their parents.
This work also addresses longstanding concerns about “generational equity” (e.g., Preston 1984), which suggest that research and policies focused on older members of a population must do so at the expense of the young, and vice versa. We provide evidence here that this is not necessarily the case. Improvements in the schooling of one generation may have consequences for others in the broader family network: in this case, older parents. Similar mechanisms to those linking one’s own schooling to improved longevity—such as greater access to health information, more income, more flexible jobs, and a healthier lifestyle—may be provided to parents by their well-educated offspring. This article contributes to our understanding of the potentially bidirectional flow of gains in the intergenerational transmission of education as well as how those benefits manifest in parents’ longevity.
Educational Attainment as a Family-Level Resource
An extensive body of research exists on the relationship between individual-level SES and health in a variety of contexts (for a review, see Elo 2009). Few studies, however, have considered family-level measures of SES and its relationship to mortality. We do know that parents’ socioeconomic characteristics are associated with their offspring’s mortality (Kuh et al. 2002) and that the socioeconomic resources of one member of a married couple are related to the survival of the other (Mare and Palloni 1988; Smith and Zick 1994). Much less attention, however, has been devoted to the relationship between adult offspring’s schooling and their parents’ health and mortality. Important exceptions include Zimmer and colleagues (2002, 2007), who have shown that offspring’s schooling is associated with older parents’ mortality and the severity of parents’ functional limitations in Taiwan, and also Torssander (2013), who has shown a similar relationship for parental mortality in Sweden.
Unknown, however, is whether adult offspring’s schooling is related to parental mortality in the United States. The U.S. context differs substantially from a more recently developed country (such as Taiwan) as well as from a welfare state (such as Sweden). In the United States, coresidence with offspring in older ages is relatively rare, financial help typically flows from parents to offspring (Eggebeen and Hogan 1990; Schoeni 1997), and the private costs of higher education are substantial.
Nonetheless, there is reason to expect that adult offspring’s schooling would matter for parental health in the United States. Although most intergenerational exchange in the United States flows downstream from parents to children, adult offspring—particularly daughters—are likely to provide for parents in later life (Lye 1996; McGarry 1998; Silverstein et al. 2002; Spitze and Logan 1990). In addition, although coresidence is relatively rare, adult offspring often live only a short drive from their parents, with Americans living, on average, within 25 miles of their mothers (Compton and Pollak 2009).
How Might Highly Educated Offspring Improve Their Parents’ Health?
Social networks, including family and friends, may influence health through numerous possible mechanisms (Berkman et al. 2000). Some of these mechanisms are direct, such as social support and enhanced access to material goods and resources; others are less direct and include the social influence that extends from the network’s values and norms. Any combination of these mechanisms may be the means through which adult children influence their parents’ health.
The influence of family on health is well reported. Marriage has long been identified as good for health (Lillard and Waite 1995; Seeman et al. 1993; Zick and Smith 1991), and the death of one spouse can hasten the death of the other (Elwert and Christakis 2008; Lillard and Waite 1995). It is not only spouses, moreover, who influence the health of someone else in their network. Adolescent siblings, for instance, influence each other’s smoking and drinking behaviors (Boyle et al. 2001; Rajan et al. 2003; Rende et al. 2005). Moreover, these effects do not end in adolescence. Smoking, obesity, and even happiness of one individual is correlated with the traits of others within their family and friendship network (Christakis and Fowler 2007, 2008; Fowler and Christakis 2008). Inasmuch as obesity and smoking behaviors are related to educational attainment, the educational attainments of particular family members may influence the health and survival of the family as a whole.
The Special Case of Smoking
The preceding discussion addresses several potential behavioral mechanisms through which offspring’s educational attainments translate into better health for their parents. In most of the discussion, offspring are treated as one of many possible close ties, such as spouses or good friends, who might influence health behaviors. However, when it comes to smoking, adult offspring may be particularly well equipped to influence parents’ behaviors. Smoking is a unique case of a negative health behavior that was once an activity of high status but is now concentrated among the less-educated and lower classes (Duncan et al. 1999; Escobedo and Peddicord 1996; Rogers et al. 1995). Smoking took hold in the United States primarily in the upper class, but as concerns about cigarettes emerged in the 1960s, individuals of higher SES were the first to reject the practice (Ferrence 1989). Because a smoking habit generally begins in the late teens or early 20s, changes in smoking habits may depend on a person being of a particular age when there is a change in the socioeconomic characteristics of smokers. This leads, then, to cohort differences in the link between schooling and smoking (Pampel 2005).1 If cohorts differ in the direction of the relationship between smoking and schooling, highly educated parents who came of age when smoking was high status may be more likely to have highly educated offspring who are not smokers. This generational divide may be less clear-cut for other health behaviors—such as dietary habits or the use of preventative medicine—that may be shared among all highly educated members of the same family.
Offspring may also very directly affect their parents’ health by providing care. Among elderly adults who need help, most receive it from family members, especially their spouses and adult offspring (McGarry 1998). Offspring and offspring-in-law provide more than one-third of the care that older adults receive and account for one-half of the care that elderly widows and widowers receive (McGarry 1998). As parents age, they are likely to have less money, fewer resources, and poorer health and thus to need more help. It is primarily their offspring who provide that support (Silverstein et al. 2002; Spitze and Logan 1990; Stoller 1983).
However, some offspring are better equipped to help than others. Offspring who themselves need assistance because of poor health or limited financial resources are less likely to provide for their parents (Hogan et al. 1993). Those with more education, on the other hand, have more resources and more flexible jobs, both of which may make them more likely to provide care. Some studies have found that adult offspring with a college degree are more likely to help parents (McGarry and Schoeni 1995), although findings are mixed, in part because highly educated offspring provide different types of help than their lesser-educated counterparts (Couch et al. 1999; Henretta et al. 1997). Importantly, more-educated offspring have better health themselves, and offspring with better health are not only positive influences on parental behaviors, as suggested earlier, but may also be more capable of providing support than their counterparts in poorer health (Eggebeen and Hogan 1990).
In addition to being better able to provide care to parents who need it, more-educated offspring may provide better care. Individuals with more schooling may have better access to health knowledge, in part because of greater access to and more familiarity with doctors, health research in the media, and comfort with the Internet. The Internet, in particular, is becoming an increasingly important means of obtaining health information and is therefore an obvious way that the younger generation can link the elderly to key health knowledge they might not otherwise obtain. However, a large sector of the population has limited access to this resource, and there remains a significant “digital divide” by education and income, among other factors (for a review, see DiMaggio et al. 2001).
In sum, the advantages possessed by more-educated offspring—better health, better access to information, and greater resources to provide for older parents—may make them more likely to provide care to their elderly parents than their less-educated counterparts. In addition, among offspring who care for their parents, those who have more schooling may provide better care, at least inasmuch as educational attainment is associated with greater knowledge of preventative health and more access to doctors and health technology. Highly educated offspring may also indirectly influence parents to engage in healthier behaviors by simply exposing parents to their own healthier lifestyles and practices. These are all possible mechanisms through which highly educated offspring might improve their parents’ health.
The purpose of this study is to examine the extent to which having highly educated offspring is related to parents’ health and survival, and to identify some of the factors that may begin to explain such a relationship. These analyses extend previous work on the relationship between schooling and mortality by looking at the association between offspring’s schooling and their parents’ mortality. To our knowledge, this association has not yet been examined in the U.S. context, as noted earlier. Using data from the Health and Retirement Study (HRS), we investigate whether adult offspring’s completed schooling is associated with parents’ survival—and, if so, what mechanisms explain this relationship. We explore this by examining the risk of dying as a function of offspring’s educational attainments, net of one’s own SES.
One concern in a study such as this is that net associations between offspring’s educational attainments and parents’ longevity may, despite extensive controls, be spurious, owing to omitted variables (e.g. genes, personality, and values) that may affect both parents’ mortality and their offspring’s attainments. We address this by investigating not only whether adult offspring’s educational attainment and parents’ mortality are associated but also how well-educated offspring alter their parents’ health trajectories. Pinpointing the possible causal mechanisms underlying this relationship, and thereby showing that offspring’s schooling is related to parental mortality in ways that are consistent with its having a causal effect on the latter, provides evidence that an observed relationship may not be spurious.
For instance, if highly educated offspring improve their parents’ longevity in part by altering their health behaviors, we would expect offspring’s schooling to be related to health behaviors and also be more strongly associated with preventable causes of death that have a behavioral component. If we do not find this relationship, an association between offspring’s schooling and parental mortality is less likely to reflect a causal relationship. Of course, finding such a relationship does not prove a causal interaction (for instance, unobserved personality traits might be associated with parental mortality, parental health behaviors, and offspring’s schooling), but it provides evidence that the observed relationships are consistent with the hypothesized mechanisms.
We use two methods to examine mechanisms through which highly educated offspring improve their parents’ health. First, we investigate the extent to which offspring’s schooling is related to parents’ cause-specific mortality, grouping deaths into those that are more closely linked to health behaviors and therefore more preventable, such as those resulting from smoking behaviors or excessive alcohol consumption. Second, we look directly at measures of smoking and exercise behaviors to see whether these are also related to offspring’s schooling, and assess the extent to which differences in these behaviors mediate the relationship between offspring’s schooling and parents’ survival. In the sections to come, we present the data, variables, and statistical methods used in these analyses. Finally, we conclude with the implications of this study for broader health and policy research.
We use data from the 1992–2006 waves of the Health and Retirement Study (HRS),2 an approximately nationally representative sample of individuals in the United States over the age of 51, along with their spouses or partners, if any. The study was first launched in 1992 and currently involves five birth cohorts: individuals born prior to 1924, 1924–1930, 1931–1941, 1942–1947, and 1948–1953. Data are collected biennially, with respondents providing detailed information about their education, income, and assets. Respondents also fill out child rosters in each wave with information on each child’s sex, age, and education.
The extensive information on the respondent’s own SES and the characteristics of their offspring make the HRS well suited for investigating the link between offspring’s schooling and parent’s mortality. These data are one of only a few U.S. sources of information on all offspring of elderly parents. In addition, for respondents and their spouses, the HRS includes detailed income information in each wave of the study and detailed wealth data.3
Our analysis sample is drawn from four cohorts: The Study of Assets and Health Dynamics Among the Oldest Old (AHEAD) cohort (born before 1924), the HRS cohort (born between 1931 and 1941), the Children of the Depression (CODA) cohort (born 1924–1930), and the War Babies (WB) cohort (born 1942–1947). The Early Baby Boomers (EBB) cohort (born 1948–1953), which was first collected in 2004, is not included in this analysis because there are very few deaths for this cohort by 2006. The RAND-HRS4 version of these data—a cleaned and streamlined collection of variables derived from the HRS—is the primary source of data for this study. Because the child roster data used in this analysis were not available in the RAND-HRS data set at the time of this investigation, we merged child variables to the original raw HRS data. In HRS households with two age-eligible married individuals, both individuals are treated as respondents. In households with one age-eligible member, this individual became the respondent, although the spouse was interviewed as well, regardless of age eligibility. For the purposes of this investigation, both household members are included in the analysis provided that they are aged 51 or older.
We begin with 26,988 individuals who were present in the HRS between 1992 and 2006 and were from cohorts other than the Early Baby Boomers (as discussed earlier). We exclude 71 individuals (0.3 %) younger than age 51 by the end of the study period, institutionalized, or with a weight of zero in all waves. We exclude an additional 1,225 (4.6 %) cases with missing or inconsistent information for birthday, interview date, or date of death. Because we use the RAND-HRS data (which are cleaned and imputed), we lose only 24 additional cases because of missing data on model variables (0.09 %). The final sample consists of 25,668 individuals; 17,772 households; 124,759 person-years; and 7,863 deaths.
Age of Death
We use a combination of two strategies to determine age of death. In 2006, data were merged with the National Death Index (NDI) for exact date of death. For respondents for whom this information is not available, we use reports of date of death from next living kin, which were obtained during survey tracking operations.
Respondent’s and Spouse’s SES
We include several measures of SES, including time-varying measures of total family income and respondent’s own educational attainment as a fixed covariate. We also include spouse’s educational attainment5 and allow this to vary over time in case a new spouse enters the household. This information is obtained directly from the spouses, who are typically also respondents. When spouses die, separate, or divorce, we retain the educational attainment of the most recent spouse. If the spouse is never a respondent in the study, if there is no spouse during the study period, or if educational information is simply unavailable for the spouse, spouse’s educational attainment is coded as missing/no spouse. In the models, schooling is coded as a series of dummy variables indicating: less than 12 years; 12 years; 13–15 years; and 16 or more years of schooling.
The HRS contains child rosters in each year with information on age, sex, and schooling for all of a respondent’s and a respondent’s spouse’s children, grandchildren, and children-in-law.6 To limit data to offspring who have completed their schooling, information is included for offspring aged 25 and older. Although not all adults have completed their schooling by age 25, limiting the offspring to those aged 25 and older eliminates most of those who are still in school. Child variables are time-varying, with offspring counted only upon reaching their 25th birthday.7
Because most of the families in this analysis have multiple offspring, there are many ways to measure and parameterize the completed schooling of offspring in the family. In preliminary analyses (not shown here), we compared several constructions of education, including whether any child has a college education, the number of offspring with a college education, the average years of schooling of all the offspring, and a measure of cumulative educational exposure8 over time. However, model comparisons suggest that the four constructions of offspring’s schooling fit the data equally well and yield similar results. In the analyses presented here, we primarily treat offspring’s schooling as a categorical variable; that is, we code offspring’s schooling as the proportion of adult offspring in the family with less than 12 years of schooling, 12 years of schooling, 13–15 years of schooling, and 16 or more years of schooling. Because these are proportions, they sum to a total of 1 for each family and allow for possible nonlinear effects of offspring’s schooling.
Cause of Death
Cause-of-death information is obtained from a match of deceased respondents in the HRS to the NDI. Our goal is to identify the causes of death most closely linked to offspring’s schooling. We estimate cause-specific Cox models for common groupings of the most prevalent causes of death in the United States in recent years (Mokdad et al. 2004), some of which are more behaviorally linked than are others. Using the International Classification of Diseases (ICD)-10, we examine nine categories of death: (1) cardiovascular disease, (2) cancers (excluding lung), (3) chronic lower respiratory diseases, (4) lung cancer, (5) accidental and violent deaths and alcohol-related disease (cirrhosis), (6) diabetes mellitus, and (7) influenza and pneumonia. All other residual deaths are captured in an “other” category (8). Finally, missing causes of death are coded as a separate category in our models (9).
We hypothesize that causes of death with a stronger behavioral component (those relating to accidental deaths and engaging in unhealthy behaviors—chronic lower respiratory disease, lung cancer, diabetes, and accidents and alcohol-related disease—will be more strongly associated with offspring’s schooling than others because educated adult offspring may help parents alter their health behaviors and avoid dangerous situations. At the individual level, the relationship between educational attainment and mortality is strongest for deaths related to known risk factors, such as smoking (Link 2008; Preston and Wang 2006), and those that are preventable and/or treatable, such as diabetes, high cholesterol, and certain cancers (Chang and Lauderdale 2009; Glied and Lleras-Muney 2008; Lutfey and Freese 2005; Phelan et al. 2010).
We also more directly assess whether parental health behaviors are related to offspring’s schooling and the extent to which they mediate the relationship between offspring’s schooling and parents’ survival. We use three dichotomous measures of health behaviors, including whether the respondent (1) currently smokes, (2) is a former smoker, and (3) is not currently engaged in vigorous exercise three or more times per week.9
The variables used for these measures come from the RAND-HRS data set, which includes variables that are coded for maximum consistency across the waves. In their first wave of entry into the study, respondents are asked, “Have you ever smoked cigarettes?” In most of the other survey waves, in addition to whether they ever smoked, respondents are also asked whether they “smoke cigarettes now.” In our models, we look at current smoking behaviors of parents. In addition, because we are especially interested in whether highly educated offspring induce their parents to quit smoking, we also use this information to classify respondents into those who had once smoked but are no longer smokers (or quit smoking) compared with those who have continued to smoke. Unfortunately, we do not have information on the precise timing of when parents quit and thus cannot assess whether it was before or after their offspring completed their schooling.
It is more difficult to capture exercise behaviors over time because the wording and response categories vary somewhat from wave to wave.10 To minimize problems resulting from inconsistencies in response categories over time (and in line with the RAND-HRS coding), we use a simple dichotomous variable to capture whether a respondent engages in vigorous exercise/physical activity three or more times per week.
Because respondents may be in any of several birth cohorts, we allow the baseline hazards to vary for different birth cohorts. We measure cohorts using a five-year grouping of year at birth. Because of small cell sizes, we group the oldest respondents into the pre-1905 cohort and the youngest respondents into the 1950 and later cohort. Baseline hazards are also allowed to vary by the stratification variables, which were used to oversample individuals with certain characteristics. These variables include dummy indicators for Florida resident at time of initial 1992 HRS sample collection and race (i.e., white, black, Hispanic, and other). Models with offspring’s schooling also include a dummy variable controlling for respondents with no offspring.
The analyses use Cox proportional hazards models (Cox 1972, 1975), which are based on ranked ages at death in days (using information on days of birth, death, and interviews). The clock is based on age in days, starting when respondents enter the study or at age 51 (whichever is later) and ending at death or censoring resulting from nonresponse or the end of the study period. Survival rates are modeled as a function of own socioeconomic characteristics, spouse’s socioeconomic characteristics, family composition, the educational attainments of their offspring, and other controls (e.g., race, region, birth cohort). Data on men and women of various ages are pooled in these analyses, yet baseline hazards are allowed to vary by sex. To prevent bias arising from nonproportionality of the survival curves, baseline hazards also vary by birth cohort, racial/ethnic group, and region. Finally, because households may include up to two respondents, standard errors are adjusted for clustering of respondents within households.
In a second set of analyses, we also run a competing risks survival analysis estimating the hazard of dying of a particular cause in the presence of multiple possible causes of death. We use a competing risk Cox model, in which each cause of death group has a unique baseline hazard and covariates interact with each of the causes.
Table 1 shows descriptive statistics calculated for the year in which respondents entered the study. Descriptive statistics are displayed stratified by respondents who remained alive or were censored between 1992 and 2006 and those who died during the period surveyed. Spouses can change over the years, as can income, wealth, and child characteristics. Time-varying covariates are collected at each wave of the study, for a maximum of eight time points. Spouse information captures the most recent spouse. Income and wealth information are measured as of the last date of interview.
Almost one-third of respondents have less than a high school diploma, just under one-third have a high school diploma, and 39 % have more than a high school diploma. Only 8 % of offspring have less than 12 years of schooling, and over one-half have at least some college. Among respondents who survived the study period, on average, 56 % of their offspring have some college or a college degree, in contrast to only 50 % of offspring for those respondents who died during the study period. This suggests that offspring’s schooling may be associated with parental mortality.
Table 2 cross-tabulates offspring’s educational attainment by the respondent’s own attainment. Although offspring and parents frequently obtain similar levels of schooling, there is a fairly wide spread in the distribution. For respondents with less than a high school diploma, for instance, an average of 18 % of their offspring have less than a high school diploma, one-half completed high school, 16 % have some college, and 16 % have a college degree or more. For respondents with college, fewer than 2 % of their offspring have less than a high school diploma, 15 % have a high school diploma, 22 % have some college, and more than 60 % complete college.
Offspring’s Schooling and Parents’ Mortality
In principle, sons and daughters may have different effects on their mothers and fathers, necessitating gender-specific analyses of parental survival. In analyses not shown here, we investigated whether there are differences in the association of offspring’s schooling and parents’ mortality by gender of parent and gender of child. These analyses show that having daughters is indeed more beneficial for mothers’ survival than for fathers’. However, we found no significant differences in the effects of sons’ and daughters’ schooling. Thus, the results presented here are based on models that do not distinguish the educational attainments of sons and daughters.
Table 3 displays the results of Cox survival models predicting the hazard of mortality. The first model shows the relationship between a respondent’s own educational attainment and survival without controls for other family members’ socioeconomic characteristics. (These models are stratified by sex, cohort, race, and Florida oversample, as described earlier.) The second model shows the association between offspring’s educational attainment and their parent’s mortality before controlling for parent’s own SES. These first two models show the associations between mortality and own and offspring’s schooling before additional controls are added. Models 3–5 are nested models demonstrating the associations between family SES and survival. Model 3 includes information on own and spouse’s schooling only, Model 4 adds offspring’s schooling, and Model 5 adds income.
Model 2 shows that with only the offspring educational variables in the model but no controls for respondent’s SES, a one-unit increase in the proportion of offspring with less than a high school diploma (i.e., going from no offspring with less than high school to all offspring with less than a high school diploma) doubles the relative odds of dying (e0.70) compared with that same increase for offspring with a college degree or more. In addition, a one-unit increase in the proportion of offspring with a high school diploma increases the risk of mortality by about 1.5 times, and even an increase in the proportion of offspring with some college significantly increases the risk of dying relative to having offspring with a college degree. These estimates also imply that in families with two offspring, if both offspring have less than a high school diploma, the relative odds of dying is double what it would have been had they completed college instead. If, on the other hand, one of these offspring has less than a high school diploma and one has a high school diploma, we see an 80 % increase in the relative odds of dying [(0.5 × e0.70) + (0.5 × e0.46)] compared with a similar family in which both offspring are college educated.
After controls for respondents’ and spouses’ schooling and family income are added (Model 5),11 the association between offspring’s schooling and parents’ survival persists, although the coefficients for offspring’s schooling are smaller. These results suggest that having more-educated offspring is associated with one’s chance of survival. In addition, these differences display a clear gradient across all levels of offspring’s educational attainment.12
Another way to examine differences in survival is in terms of median years of life lost. This can be done by comparing the median age of death for parents who have offspring with different levels of schooling. Using Model 5 from Table 3, we determine the median age of death by obtaining the underlying baseline survivor function and the predicted hazard ratios for parents for whom all their offspring have less than a high school diploma; all offspring have a high school diploma; all offspring have some college; and all offspring have college or more, with all other model variables held constant. The estimated difference in life expectancy for parents with the most-educated offspring compared with those with the least-educated offspring is about two years (age of death of 71 vs. 69). This difference is similar to the adjusted difference in life expectancy for individuals with less than a high school diploma compared with those with a college degree (difference of 1.65 years).
What Explains This Association?
In Tables 4, 5, and 6, some of the mechanisms that may explain the relationship between offspring’s schooling and parents’ survival are examined by (1) looking at cause-specific mortality to determine whether offspring’s educational attainments are related to the cause of death of their parents, specifically those which are more and less preventable; and (2) directly examining reported smoking and exercise behaviors.
Cause of Death
We investigate whether some causes of death are more strongly related to offspring’s schooling than are others. To obtain the results in Table 4, we run a competing risks survival analysis estimating the cause-specific hazard of mortality. Small sample sizes restrict us to an analysis of eight causes of death (with a ninth category for missing information).
Table 4 shows the interactions between offspring’s schooling and eight major causes of death, including a residual category. Missing causes of death are included in the models but are not displayed here. Although most of these causes of death are significantly related to offspring’s schooling, the coefficients for chronic lower respiratory ailments and lung cancer stand out from among the others. Having a greater proportion of offspring with less than a high school diploma increases the risk of dying of lung cancer or chronic respiratory disease significantly more than it increases the risk of dying of other cancers or cardiovascular disease, which are the two main causes of death for this population (and categories that include both more- and less-preventable diseases). This may suggest that there is something unique about smoking in particular and its relationship with having highly educated offspring.
A more direct way to investigate whether offspring improve their parents’ health by changing their health behaviors is to investigate whether health behaviors are predicted by offspring’s education. We estimate logistic regression models predicting the likelihood that parents (1) smoke now (vs. do not currently smoke), (2) quit smoking (vs. current smokers), and (3) do not engage in vigorous physical activity at least three times per day (vs. engage in vigorous activity). Models are clustered on individuals to account for multiple years of data. This analysis is estimated only for 24,259 individuals who provided complete information on both smoking and exercise behaviors in at least one wave. The results of these models are displayed in Table 5.
Having less-educated offspring is significantly associated with a greater likelihood that parents engage in unhealthy behaviors, such as smoking and inadequate exercise. Parents of less-educated offspring who were ever smokers are also less likely to have quit smoking compared with those of more-educated offspring. These findings show that having highly educated offspring is directly related to parental health behaviors. (For a detailed investigation of the relationship between offspring’s schooling and parental smoking cessation over time, see Field and de la Roca 2005.)
To see whether differences in parents’ health behaviors between those who have more- and less-educated offspring account for the estimated effects of offspring’s schooling on parents’ mortality, we include the measures of parental health behaviors as controls in the mortality hazard model. The first model in Table 6 shows the results of the original Cox model run earlier, but this time for the subsample of 24,259 respondents for whom smoking and exercise information is available.
Model 2 includes controls for whether respondents ever smoked; Model 3, for whether they currently smoke; and Model 4, for whether the respondent is not engaged in vigorous exercise at least three times per day. Model 5 includes information on whether the respondent ever smoked, smokes now, and does not engage in regular vigorous exercise.
First, let us consider the mediating role of smoking behaviors on the relationship between a respondent’s own schooling and survival net of the schooling of other family members. We see that whether a respondent ever smoked, smokes now, or does not exercise (Models 2, 3, and 4, respectively) all reduce the magnitude of the coefficients for educational attainment. When it comes to the relationship between offspring’s schooling and parental mortality, we see a similar trend. The inclusion of each of these health behaviors reduces the strength of the relationship between offspring’s schooling and parental mortality. Thus, health behaviors explain part of the relationship between offspring’s schooling and respondents’ mortality. In contrast, health behaviors do very little to mediate the relationship between spouse’s schooling and respondent’s mortality. Educated spouses do influence their partner’s health, but it is through some other mechanism. Admittedly, the coefficients for offspring’s schooling remain statistically significant in the last model, even after we control for parents’ smoking and exercise behaviors. This is not surprising given that there are many other mechanisms—both behavioral and relating to direct care from offspring—that are not included in these models and would explain some of this relationship. Nevertheless, the hypothesis that health behaviors are part of how offspring’s schooling is translated into survival gains for parents is supported by these findings.13
This research isolates a mechanism through which differences in health and mortality come about, to wit: the differential educational attainments of offspring. Our results suggest that in the United States, parents benefit from having more-educated offspring—a benefit that extends beyond the effects of parents’ own SES. We show that this relationship is more pronounced for deaths that are linked to behavioral factors and that may be more preventable: most notably, chronic lower respiratory disease and lung cancer. Smoking and exercise behaviors appear to be among the mechanisms that explain the relationship between offspring’s educational attainments and parents’ mortality. These findings are consistent with the hypothesis that highly educated offspring may directly improve their parents’ health by convincing them to change their health behaviors or that they may indirectly influence their parents’ health behaviors through health spillover effects. Finally, it is also possible that parents of highly educated offspring might have a stronger motivation to take care of themselves and stay healthy than those with less-successful offspring. This possibility is also consistent with our findings and warrants further investigation.14
Although health policy research typically emphasizes individual interventions with immediate outcomes, this work shows that another way to influence the health of the elderly is through their offspring. Policies targeting one generation of the family may set in motion a series of reactions that lead to improved health for others in previous generations, subsequent generations, and the broader family unit. In order to harness the value of schooling for health, then, a broader family and multigenerational perspective on health is needed. Policies with the goal of improving population health, or that of its most vulnerable members, need to consider who is best to target for the widest gains. To that end, this article has significant implications for the amelioration of health disparities, particularly unanticipated disparities resulting from kin and intergenerational ties.
These findings also relate to long-standing concerns about the possible conflict between social policies that benefit the elderly and those that benefit their offspring (Preston 1984). This article shows that generations of families are interdependent, and the well-being of one generation does not necessarily come at the expense of the well-being of other generations. Improving offspring’s lives may benefit not only the offspring themselves over their lifetimes but their parents as well.
Understanding the broader returns of investing in offspring’s schooling is particularly relevant in the U.S. context where the majority of parents help fund their children’s schooling. College educations are costly and on the rise in the United States, and parents bear much of the financial burden of high tuition costs. In fact, one study using data from the 1990s found that more than one-half of parents paid for at least some of their offspring’s schooling, with those who give spending more than $28,000 for the education of their offspring (Brown et al. 2006). Of course, in the U.S. context where healthcare in later life is also an expense incurred at least in part by the public, there is a tradeoff between investing in offspring’s schooling and holding onto the wealth and investing in one’s later-life health. Further research should consider the relative returns on this investment for the family unit.
In the United States, as well as in many other countries, there is evidence that adult offspring are critical caregivers of older parents who need help, even when they do not coreside (Bonsang 2009; Giles and Mu 2007; McGarry 1998). In addition to providing needed care to older adults, family members are crucial for cognitive, physical, and psychological well-being at older ages (Luo et al. 2012; Seeman et al. 2001). Our work suggests that the education of family members is another important factor linked to the health and longevity of older adults. This work also builds on literature showing that parental SES is an important determinant of early-life health, and one that has lasting consequences for health and mortality (Hamil-Luker and O’Rand 2007; Hayward and Gorman 2004; Kuh et al. 2002). We show that later in life, adult offspring become critical for ensuring the health and survival of their parents. This suggests that family SES matters for health throughout lifetimes, although different family members might matter more at different life stages.
It may seem surprising that we do not find statistically significant differences by schooling of parent. It is possible that less-educated parents do benefit substantially from their highly educated offspring, but the most-educated parents are better able to take advantage of this resource and are quicker to adopt new practices suggested by or observed among their offspring. The possibility that more-educated parents can better access the advantages stemming from offspring’s schooling than can others already has support in the literature. In Taiwan, for example, highly educated parents with highly educated offspring live significantly longer than lesser-educated parents with equally educated offspring (Zimmer et al. 2007).
It is worth noting that because of secular increases in completed schooling in the United States, offspring in this study are typically more educated than their parents. It is difficult to know how this might affect these findings. Perhaps education has a different meaning among parents than among their offspring. Having offspring with more education than their parents, for instance, might be less meaningful than having offspring who completed particular educational credentials, such as obtaining a high school diploma or college degree. This discrepancy might explain why we do not find stronger effects of offspring’s schooling for less-educated parents than for those with more schooling, as we had hypothesized. For the most part, however, these findings point to an independent effect of offspring’s educational attainments for parents’ mortality.
Although we focus specifically on smoking and exercise as two health behaviors that may explain the relationship between offspring’s schooling and parental longevity, an abundance of health behaviors (and other factors) may be at work. Adult offspring have the potential to influence all kinds of behaviors that are linked to overall health, including their parents’ diet, alcohol intake, medication adherence, and even time spent in cognitively stimulating activities. We focus on smoking and exercise in part because of data availability but also because these factors capture two important sides of health behaviors. Nonsmoking captures a lack of engagement in an adverse behavior, and exercise is a positive act. Our consistent findings across these different measures suggest that offspring’s influence extends across both negative and positive health behaviors. More work is still needed, however, to investigate the many other health behaviors that might similarly be related to offspring’s schooling.
A notable challenge faced by this and related studies is omitted variable bias. It is difficult to investigate the influence of offspring on parents, given that offspring’s schooling is likely correlated with family characteristics that are also related to parents’ health and risk of death. For instance, what looks like a relationship between offspring’s schooling and parental health and survival may actually be a function of some other characteristics of parents that is related to both offspring’s schooling and parental mortality. We deal with this concern by including extensive controls in these models for parents’ own characteristics and, especially, their SES. We are encouraged by the work of Torssander (2013), which shows a significant relationship between offspring’s schooling and parental mortality in Sweden in models using sibling fixed effects at the parent level. These family fixed-effects models adjust for the characteristics that parents share with their siblings, such as shared childhood experiences, abilities, personality, and values. Although the U.S. context may differ and there may be other omitted variables that remain unaccounted for in models such as these, the results in Torssander (2013) provide at least some support for our findings.
Possible endogeneity of offspring’s schooling with parental mortality, however, continues to be of concern. Healthy parents may have offspring who do better in many aspects of life, including in their educational attainments. This problem is difficult to address when using conventional analytic methods. Instrumental variables are one way to parse the effect of one factor on another while removing potentially endogenous factors, although finding an appropriate instrument can be challenging. Instead, we choose to examine the causal nature of this relationship more descriptively by investigating one of our hypothesized mechanisms through which we expect offspring’s schooling to be related to parental survival: namely, parental health behaviors. These analyses show that, as hypothesized, one way highly educated offspring improve parents’ survival chances is by improving their parents’ health behaviors.
Although more work remains to investigate the causal nature of the relationship between offspring’s schooling and parental health, this study lays the foundation for future work. These findings already suggest, moreover, that to understand how differences in health and survival come about, it is necessary to begin documenting health disparities using more sophisticated measures of SES—ones that account for the joint distribution of SES across generations of the family.
The authors thank the Robert Wood Johnson Foundation Health & Society Scholars program and the UCLA Interdisciplinary Relationship Science Program sponsored by the National Science Foundation for their financial support. The authors also benefited from facilities and resources provided by the California Center for Population Research at UCLA (CCPR), which receives core support (R24-HD041022) from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD). We are grateful to Suzanne Bianchi, Jennie Brand, Arun Karlamangla, Kathleen McGarry, James Raymo, Teresa Seeman, Judith Seltzer, and Ken Smith for helpful advice as we developed the article. Previous versions of this article were presented at the 2008 Research Committee 28 Conference on Social Stratification and Mobility (RC28), Florence, Italy; at the 2009 Population Association of America (PAA), Detroit, MI; and at the 2010 American Sociological Association (ASA), Atlanta, GA.
Using data from the General Social Survey, Pampel (2005) estimated that among whites, the percentage of ever having smoked peaks for cohorts born 1923–1938. This corresponds well to the main HRS cohort in this analysis: those born 1931–1941. Indeed, a large portion of our sample was born during the early era of smoking. The HRS sample used in this article has a median year of birth of 1934, and more than 45 % of the parents in our sample were born between 1923 and 1938.
We use “HRS” to refer to all Health and Retirement Study (HRS) cohorts.
Although we consider wealth as a covariate in preliminary analyses, wealth is not included in the models depicted in this article. Parents have the choice either to invest their wealth in their offspring’s schooling and receive the benefit of highly educated offspring in later life, or to hold onto their wealth and use this money to help themselves as they see fit. Wealth is therefore endogenous to offspring’s schooling and may confound the results of this study.
This data set (RAND-HRS Data, Version J) is produced by the RAND Center for the Study of Aging, with funding from the National Institute on Aging and the Social Security Administration, Santa Monica, CA (March 2010).
We control for current spouse’s educational attainment in order to adjust for another aspect of the respondent’s current socioeconomic status. The correlation for husband’s and wife’s schooling is .60.
In these analyses, we include only respondents’ and spouses’ biological, adopted, and stepchildren.
Details regarding drops and recoding of the child roster data are available in Online Resource 1.
Another way of thinking about offspring’s education as it affects parents is in terms of three component parts: the average offspring’s education (as described herein), the number of offspring from whom parents receive educational benefit, and the number of years that parents were exposed to adult offspring. Parents who have children early will have more years of educational exposure than parents who have their children later, and there may be important benefits associated with each additional year of exposure. In addition, parents with the same average education of offspring may fare differently depending on how many adult offspring they have. In preliminary analyses, we calculated a variable that we termed “cumulative educational exposure” (CEE), which was measured as CEE = number of adult offspring × mean years of education × total years of exposure to adult offspring. However, we found that average educational attainment of offspring was the main factor guiding the association between offspring’s schooling and parental mortality, and we therefore consider only offspring’s attainments in this article.
Questions asking about vigorous activity were not asked of the 1993 AHEAD respondents. This variable is treated as missing for these respondents in that wave.
Online Resource 2 contains additional details regarding smoking and exercise question wording and availability across waves.
In results not shown here, we also included wealth as an additional family-level socioeconomic control. The results from the model with wealth are very similar to those from Model 5 with income alone.
In analyses not displayed here, we examined whether the relationship between offspring’s schooling and parental mortality varies by parents’ educational attainment and age. We found no interaction effect of offspring’s and parents’ own educational attainments. More- and less-educated parents benefit to similar degrees from having more-educated offspring. In addition, we found that the association between offspring’s schooling and mortality weakens as parents age.
In supplemental analyses, we also incorporated information on health behaviors in our cause-specific hazard models and found that smoking and exercise behaviors explain some of the relationship between offspring’s schooling and parental cause of death, especially lung cancer deaths. (See Table S3.1 in Online Resource 3.) This confirms that these health behaviors play an important role in smoking-related deaths in particular.
We thank an anonymous reviewer for bringing this to our attention.