Intervivos Transfers and Exchange
Edward C. Norton and Courtney H. Van Houtven
March 25, 2005
Abstract
Most parents divide their bequests equally among their children, while
inter-vivos transfers are usually unequal. We propose that exchange works
better for inducing inter-vivos transfers than bequests. Inter-vivos transfers can be adjusted quickly to the amount of care, are less costly than
writing a will, and can be kept secret from other family members and
the public. The results from national longitudinal data show that, as expected, if a parent gives any inter-vivos transfers, she is more likely to
give to children who provide informal care. Informal care has no effect on
the equality of intended bequests.
1
1
Introduction
Economists have long looked for evidence of exchange as a motive for bequests.
The theoretical literature on bequests argues that parents reward children who
provide informal care and attention with a larger bequest (Bernheim, Schleifer,
and Summers 1985; Bernheim 1991). The empirical evidence, however, is mixed,
with some papers finding bequests positively associated with attention (Bernheim, Schleifer, and Summers 1985; Bernheim 1991), and others not finding
evidence of exchange (Sloan and Norton 1997; Perozek 1999).
One stylized fact is that although the burden of informal caregiving falls disproportionately on some children, most bequests are divided equally. Menchik
(1980) studied probate records from 1,050 wealthy estates in Connecticut filed
between 1931—1948. Tomes (1981; 1988) used a random sample of probate
records drawn from Cleveland, Ohio in 1967. Both researchers based their test
of altruism on the idea that equal division of an estate is evidence against altruism unless all children have an identical marginal utility of income. Menchik
found that roughly 80% of the Connecticut estates were divided equally. Tomes
found that just under half of those in Cleveland were so divided. However,
Menchik (1988) suggested that Tomes was in error and the rate of equal division was over 90% when he tried to replicate the Cleveland study. There
was a fundamental difference in how the data were selected for the two studies.
Menchik (1980) used probate records as the sole source of data, while the Tomes
(1988) sample relied on individual recall. Using linked income and estate tax
returns, Wilhelm (1996) found that the majority of wealthy decedents divide
their estate equally among their children, using federal estate tax returns. Norton and Taylor (2004) found that between 70 and 83 percent of probate records
studied in North Carolina were divided equally.
Unlike bequests, inter-vivos transfers are much more likely to be divided
1
unequally among children. We find that the majority of parents who make
inter-vivos transfers do so unequally. Why should inter-vivos transfers have
such a different pattern than bequests? Recently economists have turned their
attention from bequests to inter-vivos transfers. Empirical research on intervivos transfers has looked for evidence of the same two motives presumed to
drive bequests, exchange and altruism. Altruism is when parents equalize their
children’s marginal utility of wealth by giving more bequests to their poorer
children (Wilhelm 1996; Altonji, Hayashi, and Kotlikoff 1997; Sloan, Zhang,
and Wang, 2002). Exchange is when parents give more bequests to the child or
children who provide more informal care and attention. The empirical literature
on motives for inter-vivos transfers is less extensive than for bequests motives,
but has repeatedly found evidence of altruism (Altonji, Hayashi, and Kotlikoff
1997; McGarry and Schoeni 1997; McGarry 1999). The remarkably different
pattern of giving between inter-vivos transfers and bequests–primarily unequal
or primarily equal–suggests that the strategic interaction between parents and
children may be different for these two methods of passing wealth to the next
generation.
We argue that there are compelling reasons why exchange is a common and
important motivation for inter-vivos transfers, and furthermore that exchange is
a better strategic incentive when done through transfers than through bequests.
Transfers can be adjusted quickly to the amount of informal care and attention
actually provided, are less costly than writing a will, and can be kept secret
from other family members and the public. Transfers are a more targeted, less
expensive, and far more flexible way to exchange wealth for informal care and
attention. Therefore, it is not surprising that inter-vivos transfers are common
and are often unequally divided.
We test our hypotheses by estimating whether children of parents who make
2
any transfers are more likely to receive a transfer if they provide informal care,
and also whether parents intend to divide their bequest evenly given different
contributions of informal care. Informal care provides a good test for exchange
because, unlike some measures of attention, it involves a serious commitment
on the part of children in time and effort, and varies among children within a
family. We use data from the Asset and Health Dynamics Among the OldestOld Panel Survey (AHEAD) because it has information on both parents and
children, and about transfers, intended bequests, informal care, and finances.
Informal care is endogenous because forward-thinking children will decide how
much informal care to provide simultaneously with their parents deciding on
transfers and bequests. We test and control for the endogeneity of informal
care. As expected, parents who give any transfers are more likely to give to
children who provide informal care. The model is identified through variation
in the amount of informal care provided by children in families.
Our results support exchange as a motivation for inter-vivos transfers, but
not for intended bequests. A child who helps with activities of daily living or
instrumental activities of daily living increases their probability of receiving a
transfer of at least $500 in a year by eleven to fourteen percentage points. Our
test has the advantage over previous tests of using child-level data matched
to parents’ data; this allows us to test for differences in informal care within
families.
2
Background
Exchange has long been proposed as a motivation for bequests (Bernheim,
Schleifer, and Summers 1985; Bernheim 1991). Under exchange, adult children provide informal care and attention to their elderly parents in exchange
for a larger bequest. Exchange has great intuitive appeal as a motive for be-
3
quests because elderly persons often have a need for informal care, a desire for
attention from their children, and wealth to bequeath. Hundreds of billions of
dollars are bequeathed each year in the United States (Wilhelm 1996). Adult
children provide the most informal care to elderly parents who are not married. Among married and unmarried elderly, child caregivers (41.3 percent of
all caregivers) are more numerous than spousal caregivers (23.4 percent) (Spector 2000). Moreover, the burden of providing informal care is not generally
divided equally among the children. Exchange is not the only motive for bequests; other reasons include altruism, accidental bequests through uncertainty
in life expectancy, and precautionary savings.
Inter-vivos transfers from parents to children are common, although the
exact extent is notoriously difficult to estimate. Inter-vivos transfers are believed
to be at least one-third of all intergenerational transfers (see Gale and Scholz
(1994), for a review), meaning hundreds of billions of dollars per year. In a
study of non-elderly parents and children using the 1988 Panel Survey of Income
Dynamics, Altonji and colleagues found that 20 percent of children received a
transfer of $100 or more in a year’s time, and that the mean transfer was $297
(Altonji, Hayashi, and Kotlikoff 1997). Using a nationally representative sample
of the elderly, McGarry (1999) found that 25 percent of families made a cash
transfer to at least one child. The mean transfer was $3,013 over the past ten
years. Using the same data set, AHEAD, as McGarry we find that transfers
tend to be made unequally among children, unlike bequests. Over 60 percent of
parents who made inter-vivos transfers did so unequally, meaning that at least
one child received money and at least one did not. This estimate is conservative
because we counted a transfer as equal when all children received some transfers,
not knowing if the actual size of transfers was equal.
There are several empirical studies of motives for inter-vivos transfers. Cox
4
(1987) uses data from 1979 on intergenerational households to test how the
child’s income affects both the probability and amount of transfer. Cox develops a model of altruism and exchange that focuses on the effects of income on
transfers. Although he finds evidence in support of exchange, he is not able
measure directly whether increased care and attention affects transfers. Altonji
and colleagues (1997) look at transfers to non-resident children among parents
already actively making transfers, controlling for unobserved family heterogeneity and sample selection bias. They examine transfer behavior when a child’s
income increases and a parent’s decreases by equivalent amounts. If altruism
exists, the income change should induce a parent to reduce his or her transfer to
that child by nearly the entire amount of the increase. They soundly reject the
altruism hypothesis, because, while parents increase their transfers when their
own income goes up, consistent with altruism, parents reduce their transfers by
only a few cents for each extra dollar of income their child has. They do not
examine exchange, although a more recent study by the same authors (2000)
examines time and money transfers exchanged both ways between parents and
children. They find little evidence in the 1988 PSID that time transfers are exchanged for money in either direction, or that income or wealth increases money
transfers in either direction.
McGarry (1999) develops a model that explains the observed facts that intervivos transfers are given disproportionately to poorer children, while bequests
are usually divided equally. She predicts that parents with liquidity-constrained
children and uncertainty about their permanent income will give transfers unequally, but bequests equally, and she finds empirical evidence to support this
using data from the HRS and AHEAD. In a similar study, McGarry and Schoeni
(1997) find that parents are more likely to make a financial transfer to children in
the lowest income category, and that these children receive over $300 more than
5
children in the highest income category. They also test for exchange, and find
that parents do not provide financial assistance to their children in exchange for
caregiving. They surmise that current financial transfers are investments which
parents hope will be paid off by the child when help is needed in the future.
Our study differs from theirs in several important ways. First, we develop a
conceptual framework to explain why inter-vivos transfers would be used for
exchange instead of bequests. Second, we use two waves of AHEAD data instead of one, which gives us more variation over time within families. Third, we
obtain identification from within-family variation in informal caregiving, instead
of cross sectional variation, and thereby have better controls for unobserved heterogeneity. Fourth, we control for endogeneity of informal caregiving. Fifth, we
estimate models of intended bequests.
In another related study, Henretta and colleagues (1997) use the same data
(AHEAD) to study whether past transfers increase current caregiving–the opposite relationship that we study. They also use family fixed effects models to
exploit within family variation in past transfers and current caregiving. Henretta and colleagues find a positive and significant relationship, as expected.
Informal care is a common form of exchange to elderly parents and the
most common type of long-term care for American elderly, with about 22.4 million households providing informal care today to persons over age 50 (Family
Caregiver’s Alliance 2001). The most important source of informal care, besides spouses, is children. Children who have a low opportunity cost of their
time–low-income children, children who do not work, children who are already
caring for other dependents, or children who live close by–are the most likely
providers. Daughters are more likely to have a low opportunity cost of time
than sons and are more often caregivers. Care provided ranges from custodial tasks to complex medical care, and may substitute for formal medical care
6
(Van Houtven and Norton, 2004). While 66 percent of caregivers help with
physical functioning tasks or activities of daily living (ADLs), 75 percent help
with instrumental activities of daily living such as grocery shopping or paying
bills (Family Caregiver’s Alliance 2001). Approximately 50 percent administer
medications. Pezzin and Schone (1999) found that the number of Instrumental
Activities of Daily Living (IADLs) an elderly parent has drives demand for children as caregivers, while the number of ADLs drives demand for home health
care providers as caregivers. Duration of caregiving can also vary greatly, from
several months following a hip fracture or other health shock, to many years.
The majority of caregivers provide care for 1 to 4 years (Family Caregiver’s
Alliance 2001).
Parents rarely pay children directly for their assistance. However, monetary or non-monetary transfers may depend on care provided. Non-monetary
transfers are more difficult to measure but could include transfers of belongings
such as furniture, cars, or even the title to one’s house. Monetary transfers can
include stocks, other assets, or cash transfers. Inter-vivos transfers have tax
advantages compared to bequests, although the financial incentives for giving
apply equally to all children (Poterba 2001; Bernheim, Lemke, and Scholz 2004).
Clearly, children also may engage in exchange by providing informal care for
intangible gains, not only for present or future monetary rewards through transfers or bequests. Caregiving is physically and emotionally difficult, and while
bequest amounts may be substantial, annual monetary transfers from parents
are usually much less than what a child would earn in a full-time minimum-wage
job. Some of the intangible gains of caregiving include relief that a parent can
remain in her own home as desired; satisfaction at being able to repay a parent
for all of the effort it took to raise the child; happiness associated with a closer
relationship with a parent; fulfilling different cultural norms about the way el-
7
ders should be treated; or setting an example for their children (grandchildren
of the elderly parent) for how the adult child would like to be treated eventually.
Nevertheless, exchange for monetary transfers may be vital to children. At
the margin, receiving monetary transfers may help reassure a child that they
are doing the right thing by caring for a parent. It may alleviate the stresses of
caring for a parent, the financial sacrifice of being a caregiver, or in some other
way induce a child to continue to provide care.
3
Theoretical Framework
If a parent wants to elicit more informal care and attention from one child,
is it better to use inter-vivos transfers or bequests? Although the published
theoretical literature suggests that exchange is an important motive for bequests
(Bernheim, Schleifer, and Summers 1985; Bernheim 1991), we argue that it is
an even stronger motive for inter-vivos transfers. There are three reasons why
parents are more likely to use inter-vivos transfers than bequests for exchange.
First, inter-vivos transfers provide a stronger link between the amount of
informal care and the amount of the transfer. Assume that the parent gets utility
from informal care, which depends on both inter-vivos transfers and bequest
from the parent to the child. A parent can transfer an amount commensurate
with the level of informal care each year using periodic inter-vivos transfers (up
to $10,000 per child per year without a child incurring taxes on the transfer).
Although a parent could use bequests, the marginal effect of a dollar of extra
bequest on the supply of informal care is less than for a dollar of transfer. A
dollar of bequest may be worth less because of several sources of uncertainty.
The parent could change the will on a whim, eliminating the extra bequest. The
parent could die with no assets, perhaps after a long, expensive illness. In either
of these cases, a child who invested time and effort in informal care would receive
8
no bequest (we know that there are non-pecuniary reasons to provide informal
care to parents, but these two examples show why the monetary incentives are
weaker for bequests than transfers.) In addition, if a child explicitly provides
informal care in return for an extra bequest, it is uncertain how long the parent
will live. The payoff will generally be less if the parent lives longer because
she will consume out of savings, reducing the bequest. Therefore, uncertainty
in life expectancy means that the bequest may be inversely proportional to
the amount of informal care provided. Parents who expect to live for a long
time will be even more likely to use transfers than bequests to elicit informal
care and attention. Risk-averse children have a higher marginal effect on the
supply of informal care. Furthermore, liquidity constrained children prefer cash
now to future bequests. On the basis of incentives, transfers are preferred to
extra bequests because risk-averse children will provide more informal care for a
dollar of transfer compared to the expected value of a dollar of extra bequests,
discounted for time but not risk.
Second, it is far less expensive to change the terms of an inter-vivos transfer
than to rewrite a will. A parent can increase transfers as easily as writing a
check, and decrease transfers by doing nothing. In contrast, the cost of writing,
or rewriting, a simple will with a lawyer is typically several hundred dollars, but
may be much more expensive for a will with more terms and conditions. An
additional consideration is that if the child expecting an extra bequest knows
that rewriting a will is costly, she can shirk on providing informal care and
attention and still get the extra bequest. On the basis of legal costs, the marginal
benefit of transfers on utility is higher than for the marginal benefit of extra
bequests.
Third, inter-vivos transfers to one child can be done without the knowledge
or scrutiny of other children. Bernheim and Severinov (2003) argue that parents
9
prefer to be seen as loving their children equally. Because the will is typically
revealed to all children (and to the public if the estate goes to probate court),
unequal bequests may be seen as indication of unequal love. In contrast, intervivos transfers can be made without other children knowing. The desire to be
seen as loving equally makes parents reluctant to allow small deviations from
equality from being reflected in the will. For example, an altruistic parent
who wants to give a greater bequest to one child must balance the disutility of
unequal bequests with the altruistic desire to balance marginal utility of wealth.
On the basis of wanting to be seen as loving children equally, the marginal benefit
of transfers is higher than for extra bequests.
In summary, a parent who wants to use money to increase the amount of
informal care and attention she receives should use transfers, not extra bequests.
Inter-vivos transfers have greater incentive value, cost less, and can be kept
secret from other children. This framework suggests three hypotheses about
how exchange will affect inter-vivos transfers. First, the probability of an intervivos transfer will increase when a child provides informal care. We test this
with child-level data to see if children who provide more informal care are more
likely to receive transfers. Second, the preference for transfers is stronger for
younger parents because they expect to live longer and therefore there is more
financial uncertainty with bequests. This can also be tested with child-level
data, by comparing the magnitude of the results across samples stratified by
parent’s age. Third, informal caregiving is not a significant predictor of expected
unequal bequests. We test this using household-level data.
The empirical part of this study will focus on informal care, not attention,
although conceptually either one could be part of exchange. Informal care is
easier to measure, and involves more effort and time on the part of the child.
Attention has been defined to include making phone calls and visiting, which
10
involve comparatively less time and effort.
4
Methods
The conceptual framework leads to an empirical model in which informal care
is an endogenous predictor of inter-vivos transfers. Parents offer inter-vivos
transfers to children in exchange for informal care, which is determined simultaneously with the inter-vivos transfers. We estimate the model using child-level
data to ascertain whether a sibling who provides informal care is more likely
to receive a transfer compared to a sibling who does not provide informal care.
The methods for dealing with endogeneity are described after the main model.
We test for exchange in intended bequests using household-level data. If
parents receive informal care from any child, it is very likely that they received it
from only one child because very few parents have more than one child caregiver.
Hence, the effect of informal care on intended equal division of bequests offers
a test of exchange. If parents who receive informal care are less likely to divide
their bequests equally, that is evidence in favor of exchange in bequests. For
the many reasons discussed in the theoretical framework, exchange behavior is
not expected to be as strong for bequests as for inter-vivos transfers.
The probability of an inter-vivos transfer depends on informal care, but also
on household and child characteristics. Households with higher income and
wealth will be more likely to transfer money to their children. Besides measures
of household income and wealth we control for demographics and education.
Child characteristics include education, income, age, whether a child is a step
child, number of the child’s own children, and whether the child lives more than
11
10 miles away from the parent. The general relationship posed by our model is
Pr (transferhkt ) = β 0 + β 1 informal carehkt + β 2 household ht
+β 3 childkt + feh + εhkt
where the dependent variable is the probability of an inter-vivos transfer from
household h to child k in period t, informal care is the endogenous variable
indicating informal care provided by a child, household is a vector of household
characteristics, child is a vector of child characteristics, the βs are parameters
to be estimated, fe are household-level fixed effects, and ε is the random error
term.
The intended bequest model is very similar, except that it is tested on a
household-level data set, and the dependent variable is whether the parent intends to divide the bequest equally. The general form is
Pr (equal intended bequestht ) = γ 0 + γ 1 informal careht + γ 2 household ht
+γ 3 childht + η ht
where informal care indicates care received from any child, and η is a random
error term. The vector child includes the mean (or in some cases the difference
between the maximum and the minimum) of sibling characteristics.
The main equations for both the inter-vivos transfer and bequest models
are estimated using a logit model, with either household fixed effects or robust
standard errors that control for clustering at the household level.
Prior theoretical research implies that informal care is endogenous in models
of parent’s health care utilization (Greene, 1983). If children determine their
level of informal care simultaneously with parents’ decisions to give inter-vivos
transfers, then informal care is endogenous. If there are unobservable influences
12
on the likelihood of a transfer that are positively correlated with unobservable
influences on a child’s provision of informal care, such as the strength of the
parent-child relationship, then this will exert a positive bias to the coefficient
on informal care. However, the endogeneity bias could be negative. For example,
parent’s permanent income is positively related to transfers but makes informal
care less likely, so any unobserved part of permanent income would lead to
negative bias. If informal care is endogenous, then ignoring the endogeneity will
bias the coefficient on informal care in the main equation.
We use two main approaches to control for the endogeneity in the equations to predict inter-vivos transfers, and three approaches to predict intended
bequests. First, household fixed effects control for any time-invariant householdlevel variables that are omitted and correlated with informal care (Wong, Kitayama, and Soldo 1999). Examples include unmeasured permanent income or
wealth, the strength of the parent’s relationship with her children, and propensity to plan ahead financially. A simple correlation between caregiving and
wealth shows that those who provide care are less likely to receive transfers.
The reason is that wealth drives both caregiving and probability of transfers,
but in opposite directions. As household wealth increases, the probability of
caregiving declines and the probability of transfers increases. In the lowest
wealth quintile, the probability of caregiving is 39%, and the probability of
transfers is 5%. In the highest wealth quintile, the probability of caregiving
is 5%, and the probability of transfers is 16%. Household fixed effects breaks
the confounding effect of household wealth. One reason that we restricted the
sample to children in families in which there is at least one child who provided
informal care is because any elderly household that gave no inter-vivos transfer
to any children (i.e., there is no within-household variation in the dependent
variable) falls out of the fixed-effects model. Although we tried to estimate child
13
fixed effects models, the fixed effects remove too many observations because of
lack of variation in the dependent variable over two waves. Therefore, we use
household fixed effects in both the transfer model and the bequest model. We
also tried household-year fixed effects, and found qualitatively similar results.
The household fixed effects should eliminate the endogeneity bias due to
omitted time-invariant household-level variables, but we also used lagged values of informal caregiving, which will be less endogenous than current values
by eliminating endogeneity from reverse causation. Lagged values can still be
endogenous, however, if the error terms are serially correlated (Stern 1994;
Stern 1995). Although some endogeneity from lagged values may persist either through standard endogeneity bias, or selection bias by conditioning on
the prior wave, Stern argues that these biases are likely to be small if the serial correlation is small. The correlation is less than .12 in absolute value. We
therefore run regressions using the second wave of data and caregiving from
the first wave, along with household fixed effects. The disadvantage of this is
that the sample size is smaller, leading to less precise estimates. In addition,
we used instrumental variables to instrument for informal caregiving, although
good instruments are hard to find and the household fixed effects models are
our preferred models. We used standard statistical tests to decide whether the
decrease in bias is worth the increase in variance in the instrumental variable
model (Bollen, Guilkey, and Mroz 1995).
We present both single-equation and instrumental variables models, along
with models with fixed effects and with lagged endogenous variables, for completeness. Because the instrumental variables models lead to imprecise coefficient estimates, the household fixed effects model remains our preferred model.
In addition to estimating logit models as the main equation, we also estimated linear probability models, and reran all the specification tests. The qual-
14
itative results are nearly the same for the logit and linear probability model,
and the specification tests all reach the same conclusions. In summary, we control for the endogeneity of informal care with household fixed effects, which
will sweep away endogeneity bias due to unobserved time-invariant household
factors. These important factors include permanent wealth, strength of relationships with children, and the propensity to plan ahead. Other ways to control
for any remaining bias increases the standard errors, and, as we show later, lead
to similar empirical results.
5
Data
We use the 1993 and 1995 waves of the Asset and Health Dynamics Among the
Oldest-Old Panel Survey (AHEAD), conducted by the University of Michigan
for the National Institute on Aging, National Institute of Health (University
of Michigan 1998, 1999). Wave 1 of AHEAD is a nationally representative
sample of the community-dwelling oldest-old, defined as persons born in year
1923 or earlier. In wave 2, respondents are interviewed in a nursing home if
necessary. AHEAD contains extensive information about health care utilization
of the oldest-old. In addition, there is information about a parent’s health
status, functional status, housing, employment, income, net worth, trusts, asset
transfers among family members, health and other types of insurance, child
characteristics and about child-provided informal care.
The AHEAD survey is conducted every two years. It included 8,222 elderly
respondents in 1993 and 7,027 respondents in 1995. At least one household
member must be age 70 or older, although some household members can be
less than 70 years old. We analyze data from the 1993 and 1995 surveys at the
child level for the inter-vivos transfer model and at the household level for the
bequest model.
15
We restrict the original child-level sample of households with at least two living non-resident children (including step-children). We exclude households with
fewer than two children to focus on within-family variation in informal caregiving, instead of cross-section variation. Following McGarry (1999) co-resident
children are excluded from the child-level analysis, because co-resident children
may give and receive in-kind transfers, and these transfers are of unknown value.
Excluding co-resident children avoids measurement error in both dependent and
independent variables, and makes the comparison cleaner. The child-level data
set has 25,723 non-resident children in households with at least two children.
Because we estimate models with household fixed effects, the empirical model
is identified from transfer behavior among children within the same family. In
a household fixed-effects model the sample is limited to children of parents who
gave a transfer to at least one child, but not all children. The final sample size
is 4,872. The dependent variable is whether a parent in the household gave at
least $500 in the last year to a child (see Table 1 for descriptive statistics). Ten
percent of parents of children in the sample of 25,723 gave such a monetary
transfer to at least one of their children, with more than half of children in our
final sample receiving a transfer.
At the household level the sample is composed of households with at least
two living children (including step-children), or 7,690 households. Co-resident
children must be included for the analysis of intended bequests, because a parent
may expect to divide her estate unequally if she lives with one of her children.
To exclude co-resident children would stack the deck against finding exchange.
Of the eligible households, 4,429 reported having a will and gave information
on whether they intended to split the will equally. The dependent variable
is whether or not the household intended to divide its will equally among all
children. Approximately 95 percent of households intended to divide its will
16
equally among children.
The explanatory variable of primary interest in both the transfer and bequest
models is the receipt of informal care. Our measure of informal care at the
child level is whether or not a child provided any informal care to a parent in
each wave. Three percent of children provided informal care (see Table 1); 79
percent of parents who receive informal care receive it from just one child. This
variation in caregiving within the family identifies the parameter of interest,
because we estimate models with household fixed effects. At the household
level, the measure of informal care is whether either parent in the household
received informal care from any of his or her children. About 10 percent of
parents received informal care from at least one of their children. Because we
include co-resident children in the household level analyses, we include a second
measure of informal care, being a co-resident child. Presumably, co-resident
children provide some informal caregiving to their parents, but the extent is not
known. We controlled for the endogeneity of co-resident with fixed effects, but
did not try to find instruments for this variable, because it was not the variable
of primary interest.
Variables measuring parent characteristics are joint for the household in
both the transfer model and the bequest model. Therefore, there is no control
for gender. The average minimum age of household members is about 75; we
control for the minimum age to reflect the longest expected planning horizon
between married couples. The vast majority of households are white, and half
of households are comprised of married couples (this is conditional on having
at least two nonresident living child, so the percentage of married is higher
than the general elderly population). Similarly, the average number of children
per household is 4.4 in the child-level data (this is reasonable because elderly
with less than two children were eliminated from the sample). We selected the
17
highest level of education in the parent’s household, which averaged around 12
years. About seven percent of either parents self-report being in poor health
status, and we include this following McGarry (1999). The models also control
for income and wealth quartiles of all households. Again, following McGarry
(1999), we include a dummy variable for whether wealth is high enough to
exceed the threshold for taxing bequeathable wealth, namely $600,000 if single
and $1,200,000 if married.
The models also control for child-level variables that affect whether parents
transfer to children or plan to divide their estate equally. The average age of
adult children is about 46. We control for whether a child is a stepchild, because
the bond between step-relatives may be weaker, resulting in less transfer. About
12 percent of all of the children in the sample are stepchildren (similarly, 13 percent of all households had a step child). Each adult child has on average two
children (grandchildren of elderly persons). The average adult child has more
than a high school education and the majority lived more than ten miles away
from their parents. If parents give transfers because of altruism, then they will
try to equalize the marginal utility of their children’s income, so controlling for
children’s income and education is important. We control for child’s income because transfers may be negatively related to income if the parents are altruistic.
The child income categories changed between waves. The first seven categories
are categories from wave 1, while the others are from wave 2. The most common
income group was income between 30,000 to 50,000 dollars or to have a missing
value on income. Also in wave 2, imputed income values were used for children
if their income values were missing. Imputed income was not available in wave
1, so we created a dummy variable for income missing in wave 1. Differences
among children in their age or education could also explain altruistic behavior
by parents in giving bequests. Therefore, following McGarry (1999) we control
18
for the highest absolute difference in ages and education among children in the
bequest model.
There is one set of instruments for each level of the analysis. The child-level
instruments predict whether the child provides informal care, while also being
uncorrelated with receiving inter-vivos transfers. In many societies, the burden of caring for one’s parents falls to the eldest or to daughters, so a child’s
birth order and gender may predict informal caregiving. In addition, the child’s
other family responsibilities may compete for time, so marital status may predict informal caregiving. In the child-level sample, half were daughters, and
approximately 70 percent were married. We created dummy variables indicating whether the daughter was the eldest, youngest, or a middle sibling (if more
than two siblings). We interacted these three dummy variables with marital
status, for three more instruments.
The household-level instruments predict which parents are likely to receive
informal caregiving from their children, but are uncorrelated with transfers to
children. Health shocks and conditions are good predictors of health care utilization, including informal care, so we created ten instruments for whether either
parent (if married) reported having a problem with angina, arthritis, cancer,
vision, a major fall, hip fracture, stroke, number of ADLs, number of IADLs,
and days in bed in the last month. Because the 1993 wave was only of the
non-institutionalized, it is important to note that the surveyed population is
somewhat biased towards healthier elderly individuals (McGarry and Schoeni
1997). Not surprisingly, in the second wave, when respondents were followed
into the nursing home, average health status declined. Health problems, per se,
should not influence the probability of equal bequests. Arthritis and falls were
the most common problems, with more than a third of households reporting
problems with each of those, and hip fracture was the least common, with only
19
four percent.
We tested whether one can reject the hypothesis that informal care is exogenous by including in the main equation both actual informal care and the
error from the first-stage regression. In the inter-vivos transfer model, the null
hypothesis of exogeneity was not rejected, with a p-value of .91 (see Table 2).
Therefore, a single-equation model is preferred as long as the instruments are
valid. In the intended bequest model, exogeneity was rejected with a p-value of
.03. In both models the instruments are individually and jointly statistically significant in the first-stage regression (p<.0001), and the partial F -statistics were
4.47 and 46.37. We passed the overidentification test (p=.12, p=.27), which did
not reject the null hypothesis that the instruments can be excluded from the
main equation. Because the bequest model also has a variable for whether at
least one child co-resides with the elderly parent, we also ran the model controlling for endogeneity of co-resident children. The specification tests all passed,
and the results are qualitatively the same.
6
Results
The results at the child level show that a child who provides informal care is
more likely to receive an inter-vivos transfer than a sibling who does not provide
informal care (see Table 3). The single-equation model, appropriate if informal
care is not endogenous, has a positive and statistically significant parameter for
informal care. The average marginal effect is 16.3 percentage points, compared
to an average probability of about .5. These results, which use variation within
families that have any transfers, show that conditional on a parent making any
transfers, children who provide informal care are more likely to receive monetary
transfers than their siblings who do not provide care. In the logit model with
household fixed effects, the average marginal effect was 11.6 percentage points
20
(the estimated coefficient is slightly larger than in the model without fixed
effects). The fixed effects model is our preferred model.
These results are robust to the other specifications designed to control for
the endogeneity of informal care. When informal care is replaced with lagged
informal care, the coefficient is statistically significant at the five percent level,
and the average marginal effect is 13.6 percentage points. Not surprisingly,
the coefficient on lagged caregiving is smaller than for current caregiving in all
models. Therefore, the household fixed effects models produce marginal effects
in the range of 11 to 14 percentage points, with coefficients that are statistically
significant.
Many of the parent characteristics are significant predictors of giving transfers, especially parents’ income and wealth. Parents who are wealthier are more
likely to transfer money, as are older white parents with fewer children. All of
the child characteristics are statistically significant at the five percent level or
stronger. Children are more likely to receive transfers if they live closer than
10 miles from their parent, are younger, have children of their own, and are
well educated. Stepchildren are far less likely to receive transfers. In support of
altruism, children with higher income are less likely to receive a transfer.
Another prediction from the theoretical framework was that the results will
be stronger for younger parents. We reran all the models on subsets of the data,
stratified by the minimum household age (results not shown in tables). The
estimated average marginal effects were uniformly larger for younger parents.
For example, for the model with household fixed effects, the average marginal
effect of informal care is 41.5 percentage points if the minimum age is between
65 and 74, but only 7.9 percentage points if the minimum age is between 75 and
84.
The second main hypothesis we tested was whether or not informal care
21
affects the probability that a parent expects her bequest to be divided equally.
Informal care was not statistically significant in any of the models (see Table
4). Interestingly, the estimated coefficients and marginal effects for informal
care are similar for the logit, logit with fixed effects, and logit with lagged
informal care. There is no evidence to support informal care leading to unequal
bequests, in contrast to our results for inter-vivos transfers. The overwhelming
majority–around 95 percent–of parents who answered the question said that
they expect their bequest to be divided equally, leaving relatively little variation
in our data. Consequently, only one covariate was statistically significant in the
models–whether there is a step child. Parents are much less likely to expect to
divide their estate equally when they have stepchildren. Another consequence
of relatively little variation in the dependent variable is the caveat that this
analysis may have low power. The lack of statistical significance was also found
in models that controlled for the endogeneity of co-resident children.
The results are robust to several changes in specification. In addition to the
results presented, we also tried slightly different measures of both the dependent
variable and the measure of informal care. The substantive results were the
same when the dichotomous measure of transfers was replaced with a continuous
measure, and when it was replaced by either transfers in the last year (greater
than $500) or in the last ten years (greater than $5,000), instead of either type of
transfer. The basic results were also not sensitive to whether informal care was
measured as a continuous or dichotomous variable. When using dichotomous
measures of either the dependent or endogenous independent variables, we tried
both logit and linear probability models, and again the results were robust.
Another specification test was to see whether the results are sensitive to
whether the elderly parent is single or married. In the theoretical framework we
do not distinguish between married and unmarried persons. It is possible that
22
the results would be different if the household-level data were for a single person
or a married person. Estate tax laws are different if there is a surviving spouse,
and care and attention from children may matter more if there is no spouse. We
ran all our models on both subsets of the data, and the results are substantively
the same for both married and unmarried parents (results not shown in tables).
The magnitude, statistical significance, and results of the specification tests lead
to the same conclusions. Overall then we do not find that marital status changes
the results.
Finally, we can compare the difference between our results and those of McGarry and Schoeni (1997), who found no association between inter-vivos transfers and informal caregiving in single cross-sectional results. We replicated their
result of no effect of informal care in a model without household-level fixed effects (N = 25, 723) that also did not condition on any inter-vivos transfers. This
null result is due in part to poor controls for income and wealth. Households
with little or no income or wealth are extremely unlikely to give transfers, and
are more likely to need informal care due to health problems (there is a long literature on the positive relationship between economic status and health status
among the elderly, see Feinstein (1993)). Analyzing all households–including
those that are unlikely to give a transfer no matter how much informal care is
provided–dilutes the effect of informal care. One way to test this is to rerun
the main models (without fixed effects) by wealth quartile. As expected, the
coefficient on informal care increases with each wealth quartile. In addition,
if wealth and income are measured with error, which is almost certainly true,
and if these financial variables are correlated with informal caregiving, which is
likely, then running cross-sectional analyses will lead to biased estimates. The
advantage of controlling for household-level fixed effects is that the models control for the permanent component of income and wealth, and focus the analysis
23
on those households where transfers are most likely.
7
Conclusions
Exchange may help explain the observed patterns of passing wealth to the next
generation of children unequally when given as an inter-vivos transfer, but almost always equally when a bequest. Explaining this phenomenon is important
given that hundreds of billions of dollars are bequeathed and transferred each
year (Wilhelm 1996). We argue that although exchange may be an important
reason for bequests, it is better applied to transfers. Transfers can provide more
targeted incentives, are less expensive to change, and can be more secretive if
parents wish to keep information from other children or the public.
Our study, which directly measures exchange as a motivation for inter-vivos
transfers, finds strong evidence that increases in the provision of informal care by
adult children to their parents increases the likelihood of receiving a monetary
transfer. We use variation within families to identify exchange. The burden
of providing informal care falls disproportionately across children, with more
daughters and unmarried children providing informal care. Given the inequality
of informal care provision, and the advantages of transfers over bequests for
exchange, it is not surprising that inter-vivos transfers are so unequally divided.
Our results are consistent with those of McGarry (1999), who argued that
altruism can explain unequal inter-vivos transfers among children but equal
bequests. Although the underlying mechanism is different, exchange is another
reason to observe the same general pattern. We do not argue against altruism as
a motivation for inter-vivos transfers, but instead point out that there is strong
evidence that exchange is an additional motivator.
24
8
References
Altonji, J. G., F..Hayashi, and L. J. Kotlikoff. 1997. Parental altruism
and inter vivos transfers: Theory and evidence. Journal of Political
Economy 105(6):1121—1166.
Altonji, J. G., F..Hayashi, and L. J. Kotlikoff. 2000. The effects of earnings and wealth on time and money transfers between parents and
children, in Andrew Masson and Georges Tapinos, eds. Sharing
the Wealth: Demographic Change and Economic Transfers between
Generations (Oxford University Press), pp. 306—357.
Bernheim, B. D. 1991. How strong are bequest motives? Evidence based
on estimates of the demand for life insurance and annuities. Journal
of Political Economy 99:899—927.
Bernheim, B. D., R. J. Lemke, and J. K. Scholz. 2004. Do estate and
gift taxes affect the timing of private transfers? Journal of Public
Economics 88(12):2617—2634.
Bernheim, B. D., A. Schleifer, and L. H. Summers. 1985. The strategic
bequest motive. Journal of Political Economy 93:1045—1076.
Bernheim, B. D. and S. Severinov. 2003. Bequests as signals: An explanation for the equal division puzzle. Journal of Political Economy
111(4):733—764.
Bollen, K. A., D. K. Guilkey, and T. A. Mroz. 1995. Binary Outcomes
and Endogenous Explanatory Variables: Tests and Solutions with
an Application to the Demand for Contraceptive Use in Tunisia.
Demography 32(1):111—131.
Cox, D. Motives for private income transfers. 1987. Journal of Political
Economy 95:508—546.
Family Caregiver’s Alliance web page. http://www.caregiver.org/factsheets/
caregiver_stats.html, Fact Sheet, Selected Caregiver Statistics,
July 11, 2001.
Feinstein, J. S. 1993. The Relationship between Socioeconomic-Status and
Health: A Review of the Literature. Milbank Quarterly 71(2):279—
322.
Gale, W. G. and J. K. Scholz. 1994. Intergenerational Transfers and the
Accumulation of Wealth. Journal of Economic Perspectives 8(4):145—
160.
Greene, V. L. 1983. Substitution Between Formally and Informally Provided Care for the Impaired Elderly in the Community. Medical Care
21(6):609—619.
Henretta, J. C., M. S. Hill, W. Li, B. J. Soldo, and D. A. Wolf. 1997.
Selection of Children to Provide Care: The Effect of earlier Parental
Transfers. The Journals of Gerontology 52B:110—119.
McGarry, K. 1999. Inter vivos transfers and intended bequests. Journal
of Public Economics 73:321—351.
25
McGarry, K. and R. F. Schoeni. 1997. Transfer behavior within the family: Results from the asset and health dynamics study. The Journals
of Gerontology Series B: Psychological and Social Sciences 52B:82—
92.
Menchik, P. L. 1990. Primogeniture, equal sharing and the U.S. distribution of wealth. Quarterly Journal of Economics 94:299—316.
Perozek, M. G. 1999. A reexamination of the strategic bequest motive.
Journal of Political Economy 106:423—445.
Pezzin, L. E. and B. S. Schone. 1999. Intergenerational household formation, female labor supply and informal caregiving - A bargaining
approach. Journal of Human Resources 34 (3):475—503.
Poterba, J. 2001. Estate and gift taxes and incentives for inter vivos giving
in the US. Journal of Public Economics 79:237—264.
Sloan, F. A. and E. C. Norton. 1997. Adverse selection, bequests, crowding out, and private demand for insurance: Evidence from the longterm care insurance market. Journal of Risk and Uncertainty 15:201—
219.
Sloan, F. A., H. H. Zhang, and J. Wang. 2002. Upstream Intergenerational
Transfers. Southern Economic Journal 69(2):363—380.
Spector, W. D., J.A. Fleishman, L. E. Pezzin, and B. C. Spillman. 2000.
The Characteristics of Long-term Care Users, AHRQ publication 000049, Agency for Healthcare Research and Policy, Rockville, MD.
Stern, S. 1994. Two Dynamic Discrete Choice Estimation Problems and
Simulation Method Solutions. The Review of Economics and Statistics, 76(4):695—702.
Stern, S. 1995. Estimating Family Long-Term Care Decisions in the Presence of Endogenous Child Characteristics. Journal of Human Resources 30(3):551—580.
Tomes, N. 1981. The Family, Inheritance and the Intergenerational Transmission of Inequality. Journal of Political Economy 89:929—958.
University of Michigan, Institute for Survey Research. 1998. AHEAD
Documentation, Wave 1, University of Michigan.
University of Michigan, Institute for Survey Research. 2000. HRS 1998
Documentation, University of Michigan.
Van Houtven, C.H., and E.C. Norton. 2004. Informal care
and health care use of older adults. Journal of Health Economics
23(6):1159-1180.
Wilhelm, M. O. 1996. Bequest behavior and the effect of heirs’ earnings:
Testing the altruistic model of bequests. American Economic Review
86:874—892.
Wong, R., K. E. Kitayama, and B. J. Soldo. 1999. Ethnic Differences in
Time Transfers from Adult Children to Elderly Parents: Unobserved
Heterogeneity across Families? Research on Aging 21(4):144—175.
26
Table 1
Descriptive Statistics for 1993 and 1995 AHEAD Data
Child Data
Household Data
Variables
Mean
Std. Dev.
Mean
Std. Dev.
Dependent or Endogenous
Inter-vivos transfer
.523
.500
Plan to divide estate equally
.951
.216
Informal care
.031
.173
.105
.307
Co-resident
.140
.347
Household characteristics
Age
Non-white
Married
Number of children
Education
Poor health status
Bequeathable wealth
Respondent in Wave 2
Child characteristics
Age
Age difference
Step child
Number of children
Child’s education (years)
Education difference
Live > 10 miles away
Household instruments
Angina/chest pain recently
Arthritis in past year
Cancer ever
Eyesight (1= exc. 6=blind)
Major fall in past year
Hip fracture ever
Stroke ever
Number of ADLs
Number of IADLs
Bed days in last month
Child instruments
Eldest daughter
Middle daughter
Youngest daughter
Eldest daughter, married
Middle daughter, married
Youngest daughter, married
Married
N
Min.
Max.
0
0
0
0
1
1
1
1
74.71
.159
.499
4.41
11.84
.070
.056
.386
7.29
.365
.500
2.37
3.70
.256
.231
.487
75.81
.052
.515
3.35
12.19
.058
.054
.486
6.89
.221
.500
1.67
3.11
.233
.227
.500
38
0
0
2
0
0
0
0
105
1
1
15
17
1
1
1
46.48
8.73
.118
2.07
13.43
.323
1.61
3.29
.677
.468
47.65
8.72
.129
2.01
13.86
2.73
.652
7.25
6.18
.335
1.03
2.15
3.00
.321
18
0
0
0
0
0
0
79
51
1
9
17
17
1
.121
.462
.222
3.14
.360
.040
.138
.718
.610
1.33
.327
.499
.416
1.06
.480
.195
.345
1.39
1.19
4.85
0
0
0
1
0
0
0
0
0
0
1
1
1
6
1
1
1
6
5
31
0
0
0
0
0
0
0
1
1
1
1
1
1
1
.150
.200
.150
.105
.134
.097
.703
4,872
.357
.400
.357
.307
.341
.297
.457
4,429
27
Table 2
Specification Tests
Test
Test Statistic
p-value
Inter-vivos Transfer, child-level, if parent gave transfer
Endogeneity
χ2 (1) = .01
.91
.12
Overidentification
χ2 (7) = 11.34
Instruments correlated
F (7, 1322) = 4.47
<.0001
with endogenous variable
Intended Bequest, household-level
Endogeneity
χ2 (1) = 4.61
Overidentification
χ2 (10) = 12.20
Instruments correlated
F (10, 4385) = 46.37
with endogenous variable
28
.03
.27
<.0001
Conclusion
Not endogenous
Good instruments
Good instruments
Endogenous
Good instruments
Good instruments
Table 3
Logit Regression Results to Predict Inter-vivos Transfers, Conditional on at Least One
Transfer, Child-level Data
Logit Models
FE & Lagged
Variable
Basic
FE
Informal Care
IV
Constant
—2.10 **
—2.12 **
(.62)
(.67)
Endogenous
Informal Care
Household Characteristics
Age
Non-white
Married
Number of children
Education
Poor health status
Income & wealth quartiles
Bequeathable wealth
Wave 2
Child characteristics
Age
Step child
Number of children
Education (years)
Live > 10 miles away
Income
N
.77 **
(.22)
[.163]
1.10 **
(.28)
[.116]
.0421 **
(.0083)
—.35 **
(.14)
.126
(.097)
—.204 **
(.027)
—.015
(.016)
.01
(.19)
6 variables
.31
(.21)
—.31
(.19)
.073
(.058)
.5
(1.3)
6 variables
—.29
(.64)
—1.17 **
(.32)
—.0113 *
(.0054)
—.62 **
(.15)
.051 *
(.023)
.068 **
(.013)
—.281 **
(.073)
11 variables
4,872
—.0189 **
(.0073)
—1.25 **
(.20)
.109 **
(.030)
.085 **
(.020)
—.18
(.10)
11 variables
3,530
.87 *
(.44)
[.136]
.4
(2.1)
[.087]
.043 **
(.011)
—.35 **
(.14)
.127
(.097)
—.205 **
(.028)
—.015
(.016)
.04
(.25)
6 variables
.31
(.22)
—.30
(.19)
—.74
(.76)
.017
(.012)
—1.45 **
(.47)
—.185 **
(.055)
.113 **
(.032)
.16
(.19)
4 variables
1,119
—.0115 *
(.0055)
—.61 **
(.15)
.049 **
(.024)
.068 **
(.013)
—.30 *
(.14)
11 variables
4,872
Robust standard errors are in parentheses. Marginal effects are in brackets.
Statistical significance at the 5 percent level (*) or the 1 percent level (**).
29
Table 4
Logit Regression Results to Predict Equal Intended Bequest, Household-level
Data
Logit Models
FE & Lagged
Variable
Plain
FE
Informal Care IV
Constant
3.7 **
4.3 *
3.3 **
(1.2)
(1.8)
(1.2)
Endogenous
Informal Care
Co-resident
.33
(.27)
[.017]
—.26
(.22)
Household Characteristics
Age
.40
(.91)
[.069]
—.1
(1.2)
—.010
—.36
(.018)
(.29)
Non-white
—.26
(.29)
Married
.04
—2.7 *
(.17)
(1.3)
Number of children
.076
(.056)
Education
—.041
(.031)
Poor health status
.33
2.9
(.37)
(1.6)
Income & wealth quartiles 6 variables 6 variables
Bequeathable wealth
—.03
2.0
(.35)
(1.9)
Wave 2
.17
—.0
(.43)
(2.2)
Table 4 continued on next page.
30
.42
(.44)
[.021]
—.40
(.31)
—.55
(.47)
[—.028]
—.16
(.22)
.004
(.027)
—.42
(.39)
.08
(.26)
.086
(.083)
—.072
(.044)
.21
(.49)
6 variables
1.02
(.64)
—.003
(.018)
—.25
(.28)
.07
(.17)
.081
(.056)
—.048
(.031)
.50
(.38)
6 variables
—.01
(.35)
.08
(.43)
Variable
Child characteristics
Age
Age difference
Step child
Number of children
Education (years)
Education difference
Live > 10 miles away
Income
N
Table 4 (Cont.)
Logit Models
FE & Lagged
Plain
FE
Informal Care
IV
—.015
(.018)
—.014
(.015)
—.89 **
(.19)
.079
(.082)
.053
(.038)
—.031
(.025)
—.06
(.21)
11 variables
4,429
—.011
(.018)
—.014
(.015)
—.88 **
(.19)
.073
(.082)
.054
(.039)
—.032
(.025)
—.10
(.21)
11 variables
4,429
.39
(1.1)
.9
(1.1)
194
—.036
(.025)
—.029
(.020)
—.81 **
(.28)
.05
(.11)
.064
(.055)
.000
(.038)
—.12
(.30)
4 variables
2,113
Robust standard errors are in parentheses. Marginal effects are in brackets.
Statistical significance at the 5 percent level (*) or the 1 percent level (**).
31