In this article, we examine birth cohort differences in parents’ provision of monetary help to adult children with particular focus on the extent to which cohort differences in family structure and the transition to adulthood influence these changes. Using data from the Health and Retirement Study from 1994 to 2010, we compare financial help to children of three respondent cohorts as the parents in these birth cohorts from ages 53–58 to 57–62. We find that transfers to children have increased among more recent cohorts. Two trends—declining family size and children’s delay in marriage—account for part of the increase across cohorts. However, other trends, such as the increase in the number of stepchildren and increasing child’s income level, tend to decrease the observed cohort trend.
In this article, we examine birth cohort differences in parents’ provision of monetary help to adult children with particular focus on the extent to which cohort differences in family structure and the transition to adulthood influence these changes. A number of macro-level changes across successive cohorts have altered the American family (Furstenberg 2014; Seltzer and Bianchi 2013), including lower fertility (Morgan 1991), a later age at childbearing (Kirmeyer and Hamilton 2011; National Center for Health Statistics 2012), and increases in the odds of divorce (Cherlin 2009) as well as remarriage after divorce (Schoen and Standish 2001). The result is an increase in the diversity of families and family households (Teachman et al. 2000) as well as an increase in the prevalence of blended families (Bjorklund et al. 2007).
The transition to adulthood has also changed in important ways. For the children of the cohorts examined here—namely, children born in the late 1950s and later—the proportion of married 18- to 24-year-olds has declined (Goldscheider 1997; Payne 2011). Changes in education have led to an increase in the proportion of college graduates in the population aged 25–29 (Bauman 2016). However, young adults are more financially dependent on their parents than in the past (Kahn et al. 2013; Schoeni and Ross 2005; Wightman et al. 2013). To explore whether these cohort differences are associated with secular change in the level of parent-to-child money transfers, we first discuss the significance and determinants of intergenerational transfers and then link changing transfer patterns to cohort change in families.
Patterns of intergenerational transfers reflect the age-based structure of life course consumption and production (Lee 2003, 2012) and indicate the relative dependence of one cohort on another within a specific institutional and cultural context (Kohli 2004). Although both public and private transfers have this characteristic, private transfers from parents to adult children are additionally important because they are a component of the private cost of having children (Lee 2012). Moreover, parental money transfers to children are a mechanism for transmitting advantage across generations of a family (Albertini and Radl 2012). Parental transfers to children are embedded in the parent-child relationship, and efforts to understand them have focused on the characteristics of both parent and child.
Economic and sociological perspectives on intergenerational private transfers have traditionally differed, with the former motivated by an attempt to distinguish between altruism and exchange as donor motivations (e.g., Laitner 1997) and the latter motivated by an interest in understanding the foundations of small group solidarity (e.g., Silverstein and Bengtson 1997). Economists attempting to understand how behavior may change under different institutional conditions typically consider two motives (Kohli 2004): (1) altruism, in which the recipient’s well-being is valued by the donor; and (2) exchange, in which donor and recipient both perceive a benefit (Laitner 1997; Pezzin and Schone 1999). Although sociologists also view transfers as occurring in a particular cultural and institutional context, their primary focus has been on transfers as an integral part of small group processes in which exchange binds the group; in this framework, exchange and concern for other group members reinforce each other (Lawler et al. 2000; Mauss 1954). However, sociologists have typically been more interested in emotional exchange rather than material exchange (Kohli 2004).
Despite the varying conceptual frameworks for studying transfers, little difference exists between disciplines in the framework of outcomes and predictors used to model transfers from parent to child. Understanding transfers requires consideration of the resources, needs, and characteristics of both generations. Parental resources define their budget constraint, and children’s resources are a component of their need. Transfers may occur in the absence of a child’s need if there is at least some benefit to the donor or to group solidarity, depending on the conceptual framework. Yet, a child’s need is an important determinant of transfers because greater need is likely to increase the value of the transfer to the recipient, thereby providing a greater benefit to the donor by increasing the altruistic reward, producing a more highly valued exchange between the parties, or binding family members more closely.
Parental resources include income and assets (McGarry and Schoeni 1995; Wightman et al. 2012) as well as better parental health (Albertini et al. 2007; Brandt and Deindl 2013). A larger number of children in a family diffuses parents’ resources; thus, any one adult child in a larger family is less likely to receive help from a parent than one in a smaller family (McGarry and Schoeni 1995). Children with fewer resources and greater need are more likely to receive help from parents where need may be defined as income (Altonji et al. 1997; McGarry and Schoeni 1995) or low levels of employment (Brandt and Deindl 2013). Two characteristics that are likely to reflect parental perception of a child’s need and dependence—younger age and being unmarried—are positively associated with receipt of monetary help (Brandt and Deindl 2013; McGarry and Schoeni 1995). Family cohesion may also affect transfers. Presence of grandchildren, reflecting either greater need for assistance or a reward for continuing the family line, is positively associated with transfers (Brandt and Deindl 2013; McGarry and Schoeni 1995). Stepchildren, particularly stepchildren of the wife, receive less financial support from their parents than do biological children of both (Berry 2008; Henretta et al. 2014).
Cohort as a Link Between Family Structure and Intergenerational Transfers
Given the role of parent and child characteristics in affecting transfers, one approach to understanding the secular change in transfers is to focus on the role of cohort changes in these characteristics. Demographers have been interested in the role of cohorts in linking macro-level social change to individual lives ever since Ryder’s (1965) classic essay on the topic. Although not all social change is captured by cohort differences, cohort differences arise as members of successive birth cohorts experience key age-structured life course events in different historical periods and circumstances. Cohort differences are measured at the aggregate level, but they can be examined at the individual level because birth cohorts are simply collections of individuals who share the characteristic of period born. Thus, cohort characteristics, such as mean family size or family structure, can be conceptualized and measured as the individual characteristics of cohort members. This approach allows both inter- and intracohort variation in those characteristics, thus addressing the complex relationship of social change and cohort succession.
Combining the macro-level family changes previously discussed and the results of existing individual-level research on intergenerational transfers leads to several expectations for how a changing family structure has produced cohort changes in the dependence of young people on financial help from their parents.
Based on existing research (Kahn et al. 2013; Schoeni and Ross 2005; Wightman et al. 2013), we expect increasing levels of support of adult children among more recent parental cohorts. We expect this changing level of support to be associated with a number of cohort changes in needs and resources. Smaller family size produced by lower parental fertility, younger children at a given parental age produced by later parental childbearing, and more unmarried children produced by lower marriage rates in the children’s generation should all produce higher levels of transfers in successive cohorts. Hence, adjusting for these characteristics is expected to reduce the residual cohort difference. On the other hand, more stepchildren as a result of remarriage, fewer grandchildren as a result of marriage and fertility delay in the children’s generation, and higher educational levels leading to higher income among children are expected to reduce transfers among more recent cohorts. Therefore, unadjusted cohort differences will be smaller than when these variables are held constant statistically.
Wightman et al. (2013) examined this issue from the perspective of birth cohorts of children, finding that support for young adult children has increased over time and linking this increase to three aspects of the changing transition to adulthood: an increase in length of education, a decline in the proportion of young people married, and lower levels of full-time employment. They found no evidence that the relationship between these characteristics and support for children has changed over time.
The Wightman et al. (2013) research focused on the extent to which social change has resulted from the succession of birth cohorts of children. We examine the issue of cohort succession in the cohort of parents. The two approaches are complementary but distinct. In particular, focus on the parental cohort examines cohort change in respondent characteristics tying those respondent characteristics to defined parental birth cohorts. On the other hand, Wightman et al. (2013) tied children’s characteristics to defined child cohorts. In each case, characteristics of the other generation in the family come from multiple birth cohorts.
A second important difference is that Wightman et al. (2013) used data from a survey of young adults. The only parental characteristics included are parental education and number of parents present while the young person was in high school. Moreover, the measure of parental support is the young person’s estimate of the proportion of total support provided by the parents. Their data source precluded examination of the effects of changing fertility level (i.e., number of siblings) and other parental characteristics. Our data source, the Health and Retirement Study (described in the following section), provides richer data on both parents and children.
The Health and Retirement Study (HRS) is a biennial panel survey of the U.S. population aged 50 and older who are noninstitutionalized when they originally enter the study. The HRS maintains this coverage by introducing refresher cohorts periodically. The survey collects information from both the respondent in a household and that person’s spouse or partner, covering three domains: health, economic status, and family structure and transfers. The family data are particularly rich and include data on the characteristics of each child and stepchild as well as dollar amounts of financial transfers to and from each child. The parents’ data can be linked to each child (Soldo and Hill 1995).1
Data from the HRS allow comparisons of monetary help provided to children aged 18 and older across three parental cohorts: the portion of the original cohort (HRS) born in 1936–1941, War Babies (WB) born in 1942–1947, and Early Baby Boomers (EBB) born in 1948–1953. The analysis is limited to members of these cohorts who are in opposite-sex married or partnered households with at least one child. Birth cohort is an individual characteristic, but transfers to children are a household characteristic, and households may comprise parents from different cohorts. To address this issue, we exclude respondents married to persons outside the three cohorts so that cohorts are clearly defined, define cohort membership by the older member of the marriage, and include the difference between the ages of the husband and wife as a covariate.
Each cohort is observed over three waves, constituting two intervals, as the older member ages from 53–58 to 57–62 years. This design is illustrated in Table 1. The various waves used include all waves between 1994 and 2010. For example, the EBB cohort is examined over two intervals. In the first interval, covariates are measured at the beginning of the interval in 2006 when the cohort members were aged 53–58, and the outcome is measured at the end of the interval in 2008 at ages 55–60. In the second interval, covariates come from 2008, and the outcome comes from 2010 at ages 57–62. This design allows examination of the three cohorts at the same ages. The use of two intervals allows examination of cohort change over a wider age range. In addition, observation over a longer period allows for the episodic nature of help to adult children. The unit of analysis is the household-child dyad. There are multiple children per household. In addition, as shown in Table 1, each household-child dyad is observed up to two times. Of the 15,711 household-child dyads, 7,069 are observed over both intervals, and 1,573 are present in only one interval.
The outcome measure is the amount of transfers from the parental household to a particular child in the past two years if it exceeds $500. Respondents are asked to report the amount given. Those who respond “don’t know” are asked to place the amount in one of several brackets. We use imputations provided and documented in the RAND HRS data documentation (Campbell et al. 2014; St. Clair et al. 2011). The RAND imputations are done sequentially, imputing provision of help, the bracketed amount, and then the amount. The imputations are regression-based estimates using age, age squared, education, gender, marital status, race, income, wealth, and number of children plus a random error component. We then take the observed or imputed amount and inflation-adjust it to 2010 dollars based on the consumer price index (CPI) for all urban consumers.
We include nine variables for the respondent household measured at the beginning of the first or second interval. Male earnings, female earnings, and nonhousing household assets are dollar amounts drawn from reports provided at the beginning of the interval and are based on RAND imputations. These amounts were then adjusted to 2010 dollars using the same CPI as noted earlier. We present actual dollar amounts in the univariate statistics and log the money variables in the analysis. We use nonhousing assets because they are a likely source for transfers to children. To address negative values for assets in the analysis, we divide the variable into two components: negative logged assets and positive logged assets. The sign of negative assets is changed before logging; and households are coded 0 on the logged variable if they are in the other category. Male health and female health are each categorical variables based on self-rated health. Excellent and very good are coded 1; good, fair, and poor are coded 0. In addition, we include number of living children, including both own children and stepchildren, of the household. The data set does not differentiate between biological children and adopted children. Race and ethnicity are measured with two variables: black (coded 1 versus nonblack coded 0), and Hispanic (coded 1 for Hispanic and 0 otherwise). We also code the age difference between husband and wife in years.
We include nine measures for each child. Child’s age, a reflection of parental age when the child was born as well as an indicator of the child’s independence, is coded in three categories: 18–24, 25–34, and 35 and older. In the analysis, 18–24 is the reference category. Male is coded 1 if male and 0 otherwise. College is a categorical variable coded 1 if 16 or more years of schooling and 0 otherwise. Stepchild of male respondent and stepchild of female respondent are each coded 1 for stepchildren of the reference parent and 0 otherwise. These are child measures that are also indicators of the marital behavior of the parents. Married is a categorical variable, with married equal to 1 and not married equal to 0. Has children is coded 1 if the respondent reports that the child has children and 0 otherwise. Child’s income, labeled Income 35K+, was collected using different categories in different years and was skipped for reinterviews in some years, but it is possible to code child’s income into two categories for each of the cohorts: less than $35,000 (the reference category) and more than $35,000. Because of skip patterns, the data for both intervals are taken from 1996 for the children of the HRS cohort, 2002 for the WB cohort, and 2008 for the EBB cohort. We also include a measure of child’s employment coded in three categories: no employment (the reference category), less than 30 hours per week, and 30 or more hours per week. We impute missing data on child’s income, presence of children, and child’s employment using a hot deck procedure based on a cross-classification of the two respondents’ cohort membership, parental assets, and child’s age.
We include one context variable to measure the current economic context: the unemployment rate for the civilian noninstitutionalized population aged 16 and older. For each respondent household, the measure is the unemployment level in the household’s census region for the year at the beginning of each of two intervals (Bureau of Labor Statistics 1995–2009).
Our primary model is estimated using a tobit model that we supplement with a logit model to aid in interpretation. The tobit model estimates a latent unobserved variable, and a covariate may affect the latent variable by two processes: an increase in the probability of providing help to children or an increase in the amount of help provided. It is possible to decompose the tobit coefficient into these two components (Kang 2007; McDonald and Moffitt 1980; Roncek 1992), and the decomposition is discussed further in the Results section.
Two types of clustering occur in these data: multiple children per family and multiple observations per household-child dyad. To address this issue, we use a two-way clustered tobit model as well as two-way clustered logit model to estimate our models (Petersen n.d.).2
Table 2 presents means of the variables for the first interval. The top panel presents the outcome variable by cohort. The proportion providing help increases from 25.8 % in the HRS cohort to 29.7 % in the EBB cohort. The amount given among those with a nonzero amount is roughly stable across the HRS and WB cohorts and increases in the EBB group.
Among respondent household variables in the second panel, men’s inflation-adjusted earnings are highest in the HRS cohort, and women’s earnings are highest in the WB cohort. Inflation-adjusted nonhousing household assets rise steadily, albeit modestly, across the three cohorts. Approximately 5.7 % of households have negative nonhousing assets, primarily because of “other debts,” a category including debts arising from credit cards, medical expenses, and loans.
More recent cohorts have smaller numbers of living children, declining from a mean of 3.3 in the HRS cohort to 2.8 in the EBB cohort. The average unemployment rate in the respondent household’s census division declines across the three cohorts.
Children’s characteristics in the third panel indicate clear patterns across groups. The children of more recent cohorts are more likely to be stepchildren, and fewer are married or have children. The children of more recent cohorts also have higher incomes, and more are employed. The higher income may be due to the higher proportions of children over age 35 and college graduates in the more recent cohorts, or it may be the result of the later period of observation for these cohorts. The increase in income may be exaggerated because, unlike with incomes of the parental household, it was not possible to adjust the dummy income variable for inflation over the period. Although we expected that delayed fertility of the parents would produce a younger age structure among children of more recent cohorts, the opposite is true: more recent cohorts are older at a given age of their parents. However, this finding is primarily due to the higher proportion of stepchildren in the two more recent cohorts. On average, stepchildren are 1.8 years older than biological children, and their higher proportion in the EBB cohort accounts for most of the difference in age structure between the EBB and HRS cohorts. Delayed (or lower) fertility among the children is clearly apparent in the smaller proportion of the recent cohorts with children of their own despite their older age.
Table 3 presents results of the tobit models clustered by household and child within household. Model 1 includes only the cohort dummy variables plus race, ethnicity, and the age difference between spouses. Both of the more-recent cohorts have higher levels of transfers to children compared with the HRS cohort, but the difference between the WB and EBB cohorts is not significant (z = 1.06, p = .29). We find modestly higher levels of transfers if the spouse is younger: a larger positive age difference means a younger spouse. Both blacks and Hispanics have substantially lower levels of transfers. We address interpretation of the tobit coefficients later.
Model 2 adjusts the cohort differences for respondent characteristics. Both cohort contrasts with the HRS group remain statistically significant but decline by slightly more than one-third. The contrast between the WB and EBB cohorts remains nonsignificant (z = 0.57, p = .56). Indicators for black and Hispanic are no longer significant. Logged earnings of both husband and wife as well as logged household assets each have a positive effect on provision of help to children, indicating a declining effect at higher levels of these variables in their original metric. There are two logged asset variables: one for positive assets and one for negative assets. The sign of the negative assets is changed prior to logging, so a high level of negative assets indicates highly negative assets. The positive effect on transfers of highly negative assets may indicate that negative assets are a transitory phenomenon. Good or excellent health of either wife or husband increases transfers. Number of living children has a negative effect on the amount given to any one child. In fact, number of living children is the main driver of the decline in cohort effects observed in Model 2. Number of children alone in Model 2 produces a decline approximately equal to that observed in the model, but all the other variables without number of children reduce the cohort contrasts by only about 10 %. The economic variables, however, are the ones primarily responsible for the change in the black–nonblack contrast. Both the economic and number of living children measures are responsible for the reduction in the Hispanic coefficient between Models 1 and 2.
Model 3 adds child covariates. Holding child’s characteristics constant increases the cohort contrasts compared with Model 2, although they remain smaller than in Model 1. The contrast between the WB and EBB cohorts remains nonsignificant (z = 1.24, p = .21). The coefficient for age difference between spouses is reduced to near 0. While the black–nonblack contrast remains nonsignificant, the Hispanic contrast increases and is statistically significant. Among respondent characteristics, respondent earnings and assets continue to have positive and significant effects on help provision, and the negative effect of number of living children also continues.
Among child covariates, being the stepchild of the female respondent, being older, being married, having higher income, and being employed full-time are associated with receiving lower levels of support. Having children is associated with greater support, and college graduation and gender have null effects.
Overall, the addition of child variables increases cohort contrasts compared with Model 2. Three suppressor variables account for this phenomenon. Child’s income and stepchild status are most important, and unemployment is a secondary factor adding to the effect. Child’s income is higher in more recent cohorts and particularly in the EBB cohort, and higher income reduces the level of transfers. When child’s income is held constant, larger cohort contrasts emerge because the effect of higher income on suppressing transfers has been eliminated by the statistical control. The same pattern characterizes stepchild status. In the case of unemployment, more recent cohorts have experienced periods of lower unemployment. Lower unemployment is associated with lower transfers. Holding unemployment constant, the suppression of lower unemployment on transfers in the more recent cohorts is eliminated by the statistical control, and the cohort contrast increases. In sum, when these three variables are omitted from Model 3, cohort contrasts are slightly smaller than in Model 2, although they remain statistically significant. When they are held constant, larger residual cohort contrasts emerge.
The increase in the Hispanic coefficient is caused by a large number of the children’s characteristics, but the two most important are age and income. The children of Hispanic respondents are younger and have lower income. These two characteristics are associated with greater transfers. They are not held constant in Model 2, thereby raising the negative Hispanic coefficient to close to 0 because these characteristics are associated with higher transfers. However, within levels of these variables, Hispanics receive less, and the inclusion of these variables in Model 3 releases the suppressive effect and increases this negative coefficient.
We examine interactions between cohort membership and the covariates and do not reject the null hypothesis of no difference in covariate effects among cohorts (chi square = 41.4, df = 44, p = .58).
The tobit models an underlying unobserved variable. Although the coefficient has little intuitive meaning, it is possible to decompose it into two parts: (1) the proportion due to a covariate’s increasing the probability of giving some help versus none, and (2) the proportion of the coefficient reflecting the effect of a covariate on the amount given among those who are above 0 (McDonald and Moffitt 1980; Roncek 1992). It is then possible to estimate the marginal effect of a variable given a positive amount. This marginal effect is reported in Table 4. We evaluate effects at the predicted probability of providing help produced by the mean level of the covariates in Model 3 for each cohort.3 The point of evaluation is particularly important because the decomposition depends on the location chosen (Kang 2007).
The left column in Table 4 reports the proportion of the tobit effect resulting from the marginal change above the threshold (i.e., among respondents providing help). The results are shown by cohort, and the same proportion applies to each covariate for members of that cohort. We see relatively little variation across cohorts, with values ranging from approximately 0.21 to approximately 0.23 of the tobit coefficient due to the marginal effect for those providing some help. The marginal effect for those above the limit is this decomposition fraction multiplied by the tobit coefficient (Kang 2007). The right column of Table 4 provides this marginal effect for the Model 3 cohort contrasts with the HRS cohort. Among those who provide help, the two more recent cohorts—the WB and EBB cohorts—provide an estimated $490 and $721 more help to children, respectively. These amounts are fairly small, indicating that most of the tobit coefficient reflects the difference between those who provide help and those who do not. The proportion of the effect that is due to an increase in the probability of providing any help is calculated as 1 minus the proportion presented in Table 4. Approximately 77 % to 79 % of the tobit coefficient is due to the effect of cohort and the other independent variables on the probability of providing some help versus none.
To aid in interpretation, we estimate a logit model in Table 5 that mirrors the tobit final model. The outcome in this equation is providing some money help versus none. Overall, the pattern of statistically significant results replicates the tobit model, something that should be expected given that the tobit decomposition indicated that approximately three-quarters of the tobit coefficients were due to the difference between those who provided some transfers and those who provided none.
Our analysis classifies those who marry someone in a different cohort by the age of the older spouse. We examine sensitivity of the results to this decision by conducting a parallel analysis classifying households by the cohort of the younger spouse. This reclassification changes the year of observation for those in cross-cohort marriages, implying an older average age for respondents as well as the children. The final tobit model (parallel to Model 3) is quite similar to the one reported. The cohort contrasts are larger but follow the same pattern of a decline between Models 1 and 2, followed by an increase in Model 3. Some change is evident in specific covariate estimates, but the results overall are robust to the reclassification of cross-cohort marriages.
Summary and Discussion
Changes in family structure across successive cohorts have influenced intergenerational transfer patterns by changing the relative dependence of adult children. We begin by reviewing five trends in family structure and the transition to adulthood that have this potential because they affect the needs, resources, and characteristics of parents or adult children: (1) lower fertility, implying fewer children per family; (2) later childbearing, producing younger children at a given age of the respondent; (3) divorce and remarriage, producing more stepchildren; (4) lower marriage rates in the children’s generation, producing more single persons and a lower proportion with children of their own; and (5) higher levels of education, producing higher levels of income in recent cohorts of children. We find evidence for all these trends across the three cohorts examined except for delayed fertility. Children of the two later cohorts have a similar or older age distribution compared with the HRS cohort, primarily because of the proportion of stepchildren in the two younger cohorts. We do see evidence of delayed fertility or decline in fertility in the proportion of the respondents’ children who have children. Despite being older, children of the two more-recent cohorts are less likely to have children compared with the HRS cohort.
Both the tobit and the logit results are consistent with existing micro-research on intergenerational transfers and indicate that the individual-level measures of these cohort trends affect transfers. Among respondents, larger family size reduces intergenerational transfers to any one child. Younger children are more likely to receive help. Parents’ remarriage and blended families, reflected in the child measure of stepchild status, reduces transfers, particularly to stepchildren of the female. Among the child variables, single children are more likely to receive transfers. Presence of grandchildren has a positive effect on transfer receipt. Those with higher income or full-time employment receive less in transfers.
The secular increase in transfers observed in these data can be partly understood as the outcome of the changing demographic structure. Among respondent characteristics, number of living children is a major contributor to the observed difference among cohorts because larger families diffuse parents’ resources. Smaller family size increases transfers to an individual child; adjusting for it, therefore, reduces net cohort differences. Child’s marital status has a similar effect: more recent cohorts are less likely to be married, thus increasing transfers, which accounts for a portion of the cohort effect. Hence, we conclude that cohort change in these characteristics has partially driven the increase in transfers in more recent cohorts. However, other demographic cohort changes—specifically, higher child’s income, more stepchildren, and older age of children—decrease observed cohort differences. Their higher level in more recent cohorts would be expected to decrease money help from parents. Therefore, a model that adjusts for them increases cohort differences compared with a model that does not through the suppressor mechanism discussed earlier.
Because some of the individual-level measures of family change decrease cohort differences while others increase them, it is not possible simply to look at the change in the cohort contrasts to summarize the results in a straightforward way. The overall effect of the model, however, is to reduce the cohort differences by approximately one-third between Models 1 and 2 and approximately 18 % to 25 % between Models 1 and 3.
In sum, in the cohorts examined, we discern a pattern of increasing transfers to children in successive cohorts. A portion of this change can be attributed to changing family structure, but the relationship is complex, with some variables producing an increase in cohort differences and others reducing observed cohort differences. The results reported here, strictly speaking, apply to the specific cohorts and the specific historical period examined. On the other hand, the increasing levels of transfers across cohorts have been found in others’ work (Kahn et al. 2013; Schoeni and Ross 2005; Wightman et al. 2013), suggesting that it is a characteristic of a somewhat broader historical period.
The overall significance of this research lies in two areas. First, the analysis links two levels of analysis: macro-societal changes in the family measured at the cohort level, and micro-level household decisions. Decisions that individuals typically view as personal and idiosyncratic—such as giving money help to an adult child—reflect broader social and historical changes (Mills 1959) that have formed the family structure of members of each cohort. The micro-level determinants of transfers and the macro-trends in the family are well known. The contribution of this analysis is in linking the secular changes in the family to individual behavior through the prism of cohort succession. It is particularly important to note that we found no evidence for change in the likelihood that a parent will help a child given that child’s characteristics. Rather, it is respondents’ and children’s characteristics that have changed. Second, our analysis illustrates that change across cohorts is not necessarily a simple unidirectional change. Some observed family changes across cohorts have led to increased transfers, whereas others have tended to reduce observed differences. Thus, cohort effects can be complex.
This research was supported by National Institutes of Health/National Institute on Aging grants to authors John Henretta (R01 AG 024051) and Beth Soldo (R01 AG024046).
Full documentation and data download is available from the HRS website (http://hrsonline.isr.umich.edu/).
These models are available in STATA using the tobit2 or logit2 command in tobit2.ado or logit2.ado, respectively. See Petersen (n.d.).
We also used means of categorical variables. Although it makes no sense to talk of an individual being a 0.35 college graduate, at the population level, we can think of the predicted probability of providing help in a population composed of 35 % college graduates. We estimated the decomposition for someone with positive assets (approximately 94 % of the households have positive assets).