Full text of DOI:10.1093/geronb/57.1.S23

Journal of Gerontology: SOCIAL SCIENCES
Copyright 2002 by The Gerontological Society of America
2002, Vol. 57B, No. 1, S23–S32
Whose Education Counts? The Added Impact
of Adult-Child Education on Physical Functioning
of Older Taiwanese
Zachary Zimmer,1 Albert I. Hermalin,2 and Hui-Sheng Lin3
1Population Council, Policy Research Division, New York, New York.
2Population Studies Center, University of Michigan, Ann Arbor.
3National Institute of Family Planning, Taichung, Taiwan.
Objectives. Research has implicated education as an important predictor of physical functioning in old age. Older
adults in Taiwan tend to experience tight familial integration and high rates of adult-child coresidency—much more so
than is typical in Western cultures—which might imply additional influences stemming from the education of children.
This could arise in a number of ways; for instance, through the sharing of health-related information between child and
parent, the quality of caregiving efforts, monetary assistance for medical and other services, or other psychosocial ave-
nues. Despite this probable association, such hypotheses have rarely been tested. In this study, a nationally representa-
tive survey of older Taiwanese was used to examine these concurrent effects.
Methods. Outcome variables include the existence of any functional limitations (dichotomously measured) and the
severity of functional disorders (ordinally measured). Dichotomous and ordinal logistic models were used.
Results. Results suggest that, after adjusting for age, sex, and other factors, both child and respondent education as-
sociate with the existence of limitations, but the child’s education is more important than the parent’s when predicting
severity of limitations.
Discussion. This implies that models ignoring social network characteristics in determining health outcomes of older
adults may be misspecified, at least in some non-Western societies, and calls for further testing in other societies as well.
family is an economic unit bound by emotional
ties. . . . It is in the household that the larger social
work. In the current study, we sought to answer the question
of whether the education of an adult child associates with
the physical functioning of their aged parents in addition to,
or over and above, the effects of the older adult’s own level
of education. If so, what aspects of physical functioning as-
sociate with child’s education? To answer these questions,
we examined the influence of parent and child education,
separately and combined, on the probability that an aged
parent has functional limitations and on the severity of those
limitations.
There is a lengthy social tradition that examines how
stratification affects health, although most of this has con-
cerned populations in the United States and Britain. By re-
viewing decades of empirical evidence that shows a robust
association between class and mortality, Antonovsky (1967)
helped initiate the modern discourse on the topic. His re-
view was followed by extensive studies of individual socio-
economic characteristics on health and mortality in the
United States and England (Kitagawa & Hauser, 1973; Mar-
mot, Rose, Shipley, & Hamilton, 1978). More recently, at-
tention in Western industrialized countries has turned to in-
vestigating the changing trends in socioeconomic status and
health over time (Pappas, Queen, Hadden, & Fisher, 1993;
Preston & Elo, 1995; Townsend & Davidson, 1982) and ex-
amining causal mechanisms and intervening factors that
serve to disadvantage those of lower socioeconomic status
(House, Lepkowski, Kinney, Mero, Kessler, & Herzog,
“A
and economic order impinges on individuals, exposing them
to varying degrees of hardship, frustration and struggle”
(Ross, Mirowsky, & Goldsteen, 1990, pp. 1059–1071). This
quote illustrates the principle that families interact and in-
fluence one another in a multitude of ways, and these inter-
actions serve to shape the life experiences of family mem-
bers. There is some debate over the nature of family in
American society (Luescher & Pillemer, 1998; Silverstein
& Bengston, 1997), but there are also societies in which the
existence of high degrees of family cohesion, interaction,
coresidence, and reciprocity are less debatable. Subse-
quently, the potential for interactions between families and
individuals within families is substantial across much of the
world.
The quote above notes that among the life experiences
shared and shaped by families are health outcomes. Over-
looking the influences of the family can result in limited in-
sights into the factors that determine health. The current re-
search took place in Taiwan and is born out of an extensive
history of research documenting an association between so-
cioeconomic status and health. Although there are some re-
cent exceptions (Robert & Li, 2001), research contemplat-
ing the socioeconomic determinants of health among older
adults usually involves what can be called an individualistic
perspective, ignoring the attributes of a wider social net-
S23

S24
ZIMMER ET AL.
1994; Link & Phelan, 1995; Ross & Wu, 1995; Williams,
1990). With respect to those at older ages, there is some ev-
idence of convergence in rates of mortality and chronic dis-
orders by levels of education (Elo & Preston, 1995; House
et al., 1990). Yet there is also convincing evidence that sig-
nificant differentials in a variety of physical functioning
abilities are maintained into old age (Camacho, Straw-
bridge, Cohen, & Kaplan, 1993; Duffy & MacDonald,
1990; Guralnik, Land, Blazer, Fillenbaum, & Branch, 1993;
Kaplan, Strawbridge, Camacho, & Cohen, 1993; Rogers,
Rogers, & Belanger, 1992). Such differences may reflect
the cumulative impact of education throughout life (Lynch,
Kaplan, & Shema, 1997; Ross & Wu, 1996).
Most of the studies noted above focused on the socioeco-
nomic characteristics of the indexed individual, and differ-
entials in health were explained as a function of personal
characteristics including one’s access to medical care, per-
sonal behaviors, knowledge of disease prevention, and psy-
chological dimensions such as stress and feelings of self-
efficacy. But health crises and ongoing conditions tend to
involve family networks in many ways. Adult children may
assume not only caretaking roles, but they also bring con-
siderable financial, knowledge, and management resources
into decisions about the timing and quality of health care
and subsequent actions. To the extent that this is so, studies
of the influence of socioeconomic status on health that fo-
cus solely on the individual respondent’s attributes may
misspecify the determinants of health of older adults.
havior within it. The occupation of Taiwan by the Japanese
from 1895 to 1945 did little to change the basic organization
of the family (Hermalin et al., 1996). Since that time, Tai-
wan has undergone demographic and socioeconomic changes
that have transformed it from a rural-based to an urban-
based society and from having a less developed economic
infrastructure to a more developed one. Rapid reductions in
fertility and increases in life expectancy over the last two
decades have created a rapidly aging society. Yet the cohe-
siveness of the family appears to remain intact. Household
sizes, although decreasing slightly over time, remain large
(Asis, Domingo, Knodel, & Kalyani, 1995), and intergener-
ational support remains strong across a number of economic
and caregiving domains (Hermalin, Ofstedal, & Lee, 1992).
An additional change, noteworthy for the current study, is
the rapid rise in educational attainment and the diminution
of the gender gap in education that have occurred over the
last several decades. College attendance for men 18 to 21
rose from about 2% to about 29% between 1952 and 1988,
whereas for women the increase was from almost 0% to
about 33% during the same time period (Hermalin et al.,
1996, Table 1). This has created a situation in which the
current generation of older adults has educational levels
well below those of their children. Children are now ex-
posed to a new set of beliefs and an expanding knowledge
base that may alter and expand the resources available in the
care of their aging parents (Cornman, Hermalin, Roan, &
Chang, 1996).
The Taiwan Setting
METHODS
Because of its Confucian culture, and what this means for
intergenerational support, Taiwan represents an ideal setting
for studying the impacts of an adult child’s education on
older adult health. The Chinese who originally settled in
Taiwan, beginning in the seventeenth century, brought with
them a family system based on extended households and the
notion of family as a corporate unit (Hermalin, Ofstedal, &
Chang, 1996). As a corporate unit, the Chinese family is or-
ganized around a collective economy whereby earnings are
pooled and used to provide support for all family members
(Lang, 1946). This notion extends beyond economics to
providing assistance with household maintenance and care-
giving. There are good reasons to presume that the influence
of the child on parental health would be substantial in this
type of system. A pooling of material and information re-
sources translates into a variety of health-related functions
that the family will perform. The family can provide a great
deal of personal care. This includes paying for medical
costs, assisting elders in daily activities, and providing so-
cial interaction and recreation. The physical proximity of
much of one’s social network also means that family re-
sources can be efficiently mobilized. In turn, this may make
factors such as adult children’s health-related knowledge
and their access to medical care—factors that relate to their
educational attainment—important in the health outcomes
of their older parent.
Data
The data come from The 1989 Survey of Health and Liv-
ing Status of the Elderly in Taiwan. This survey involved a
large sample of older Taiwanese, selected randomly to be
nationally representative. Adults aged 60 and older at the
end of 1988 were considered to be the sample universe. One
of every 370 older adults living in Taiwan was selected for
study, 4,049 individuals were interviewed, and the response
rate was 92%. Of the respondents, 3,795 had at least one liv-
ing child. The analysis was restricted to these individuals.
The main aim of this project was to collect information use-
ful in analyzing the influence of rapid social and economic
change on elderly adults in Taiwan. As such, the survey it-
self is extremely extensive and includes information on the
physical health and sociodemographics of the respondent,
but also includes sociodemographic information on all
coresident and noncoresident children. These qualities make
this data source ideal for comprehensive testing of the rela-
tionship between education and functional status and the
possible added influences of social network members. More
information on this survey is found in the first research re-
port published as part of the project (Taiwan Provincial In-
stitute of Family Planning, 1989).
Measuring Physical Functioning
Functional health can be defined as the ability to conduct
tasks such as activities of daily living (ADLs; Katz, Ford,
Moskowitz, Jackson, & Jaffe, 1963), instrumental activities
of daily living (IADLs; Lawton & Brody, 1969) or tasks in-
Over the past several decades, Taiwan has undergone a
number of changes that have had an effect on the demo-
graphic and socioeconomic structure of society, but have
not altered the basic nature of the family and normative be-

WHOSE EDUCATION COUNTS?
S25
volving general upper and lower body movements (Crim-
mins & Saito, 1993; Nagi, 1976). Some of these types of
tasks, particularly IADLs, can be confounded by role expec-
tations and living environments (Freedman & Martin, 1998;
Simonsick et al., 2001; Verbrugge & Jette, 1994). For in-
stance, the answer to a question such as “Do you have any
problem shopping for things?” may depend, to some degree,
on the distance one needs to travel to buy things or on who
tends to be responsible for doing the shopping. The distinc-
tion between these types of activities has been made by
Nagi (1976, 1991), who identified functional limitation, the
loss of ability to perform tasks, as a specific state mediating
impairment, or physiological abnormality, and disability, or
behavior patterns that evolve from impairment (Kelly-
Hayes, Jette, Wolf, D’Agostino, & Odell, 1992). In the cur-
rent study, we sought to limit our physical functioning mea-
sures to those that indicate limitations and therefore involve
upper or lower body movements. By focusing on limita-
tions, we sought to avoid the social biases that can relate to
disability. Specifically, we used ability crouching, climbing
stairs, walking, grasping, and reaching for things. A number
of other items in the data, such as doing housework, were
omitted because responses may reflect role expectations and
living environments as noted above. Other items, such as
lifting things and standing for a period of time, were omitted
because of a high number of missing responses. Specifi-
cally, many individuals responded that they never did these
tasks and therefore could not assess their ability.
Our conceptualization of functional limitations assumes
two distinct measures. All ailments, whether they are chronic
conditions or temporary illnesses, can be measured dichoto-
mously, indicating the existence of the ailment, or using an
ordinal level scale, indicating the severity of the ailment.
For instance, with respect to functional capacity we can first
measure the existence of any functional limitations. Second,
for those who have limitations, we can measure the severity.
Many individuals, even at older ages, are not troubled with
functional limitations at all. For those who are troubled with
limitations, the severity will vary. In terms of education, this
assumes two questions. First, how does education influence
the probability of having a functional limitation? Second,
how does education influence the severity for those who
have functional limitations? If determinants act distinctly on
the existence versus the severity of functional limitations,
then combining both effects into a single equation may dis-
tort the effect of this variable. For this reason, we con-
structed two functional outcome measures, the first indicat-
ing existence and the second indicating severity among
those who have functional limitations.
for this group of items was .89. This suggests that they rep-
resent a single construct.
Respondents who reported at least a little difficulty with
at least one of the five functional tasks were considered as
having a functional limitation and were given a code of 1
with respect to the existence of functional limitation vari-
able. Others were given a code of 0. To construct a severity
variable, we examined the specific responses to the five
items and determined whether the limitations reported were
mild, moderate, or severe. Those reporting only a little diffi-
culty with only one or two tasks were considered to have
mild limitations. Those individuals reporting “a lot of diffi-
culty” with, or that they “cannot do,” one or two functional
tasks, or those individuals reporting “a little difficulty” with
more than two (that is three to five) functional tasks, were
coded as having moderate limitations. Finally, those indi-
viduals reporting “a lot of difficulty” with, or that they “can-
not do,” three or more functional tasks were coded as hav-
ing severe limitations. Of those who reported at least one
limitation, 43% were coded as mild, 38% as moderate, and
19% as severe.
Other measures that have been used in the past to indicate
degree of severity include number of limitations reported
and a composite scale on which points are awarded for no
difficulty, some difficulty, a lot of difficulty, or being un-
able to do the task and the points are summed across avail-
able items. Although these types of measures are referred to
frequently in the literature, there has been no agreement on
one best way to measure severity, and each measure has
limitations. For instance, an individual may have a high
number of functional limitations, but if they report only a
little difficulty with each task, the total number of limita-
tions may not be reflective of severity. Similarly, points ar-
bitrarily awarded to response categories, say a score of 1 for
“a little difficulty” and a score of 2 for “a lot of difficulty,”
may not reflect a true difference in severity. It is uncertain,
in this example, that having a lot of difficulty is exactly
twice as severe as having a little difficulty. Yet, all of these
scales are highly correlated and represent a similar con-
struct. In the current data there was a .94 Pearson’s r be-
tween the number of limitations and the ordinal severity
scale, and there was a .89 correlation between a summative
score and the ordinal severity scale. Furthermore, sensitivity
tests suggest that our conclusions do not differ when using
these other scales. The choice of severity measure is then
somewhat based on preference and relation to the particular
research task. We used the ordered measure because we
were interested in capturing a qualitative degree of differ-
Table 1 provides the distribution of functional limitation
items including the possible response categories. This table
is organized from the most difficult task to the easiest ac-
cording to the proportion reporting no difficulty. The most
frequently reported difficulties were crouching, with about
27% reporting at least a little difficulty, and climbing stairs,
with about 26% reporting at least a little difficulty. On the
other hand, over 90% reported no difficulty grasping or
reaching. A factor analysis of these items (not shown in tab-
ular form) showed that they load highly onto a single factor
(eigenvalue of 3.5), and the Cronbach’s alpha coefficient
Table 1. Distribution of Difficulty With Functional Task Items
No
difficulty
A little
difficulty
A lot of
difficulty
Cannot do
the task
Task
Total
Crouching
Climbing stairs
Walking
Grasping
Reaching
73.3
73.9
80.0
91.7
92.6
12.9
15.3
10.4
4.5
6.8
5.4
5.0
1.7
1.4
6.9
5.5
4.6
2.2
2.2
100.0
100.0
100.0
100.0
100.0
3.7

S26
ZIMMER ET AL.
ence in severity while using categories that best parallel the
response categories of the original items.
der of children is important because men are thought to
have greater earning power than women, and parents may
disproportionately rely on their male offspring for financial
support and the wife of male offspring for ADL and IADL
support (Hermalin et al., 1996). In cases in which there was
just one child with the highest level of education, or more
than one child, but all with the same gender, we coded the
gender as male or female. In cases in which there was at
least one male and at least one female child with the same
high level of education, we coded the gender as both male
and female. We constructed dummy variables to test for
the influence of the gender of the child, with female being the
comparison category. In a similar vein, we measured the
residential proximity of child to parent. It is reasonable to
expect that the education of a child who is a coresident will
be more influential than that of a child living farther away.
If two or more children had equally high levels of educa-
tion, we considered the proximity as the child living closest
to their parent. Residence status for the most highly edu-
cated child or children was coded into three categories:
coresiding, living nearby (in the same community, village,
or town), or living farther away, with the last being the com-
parison.
We also considered a set of covariates that related to the
demographic characteristics of the parent and were ex-
pected to influence physical functioning outcomes. Age is
an important correlate of health, and those who were older
were expected to have a higher probability of reporting lim-
itations. We adjusted for gender of respondent because
women have been shown, in the past, to be more likely to
suffer from functional disorders at older ages (Verbrugge,
1989). Access to medical facilities may differ between rural
and urban areas, and we included rural or urban residence in
our models. Marital status has been shown to influence
health in a number of ways, such as through support (Gold-
man, Korenman, & Weinstein, 1995; Ortmeyer, 1974). We
considered marital status as a dichotomous variable—mar-
ried versus other. Taiwan is fairly homogenous ethnically,
but there is an important distinction between “Mainlanders”
and others. Mainlanders are those who migrated to Taiwan
from China after the Second World War. They are predomi-
nantly male soldiers and officials and have distinct social
and behavioral characteristics that warrant particular atten-
tion (Hermalin et al., 1996). For instance, Mainlanders are
more likely to be employed in the government, to be unmar-
ried, and to have higher incomes than are other Taiwanese.
Because they were soldiers, Mainlanders may have higher
fitness levels than do others, and they may consequently
have fewer functional limitations. Our ethnicity variable
therefore contrasted Mainlanders with others. We also in-
cluded a measure of total number of children in our model
to account for the overall availability of support from chil-
dren.
Covariates
Education has been shown to relate to knowledge and un-
derstanding of disease processes, preventions and practices,
self-esteem, and sense of control. It is a consistent measure
of socioeconomic status, remaining constant throughout
adulthood for most individuals. It is also a relatively acces-
sible and reliable measure. Individuals not only know their
own level of education, but most also can accurately report
their children’s. This is in contrast to, say, income, which
may be accrued in a variety of ways, changes throughout
life, and has high levels of reporting error, particularly on
reports of the income of family members other than the re-
spondent. These considerations have led a number of au-
thors to consider education the aspect of socioeconomic sta-
tus most important to health (Ross & Wu, 1995; Winkleby,
Jatulis, Frank, & Fortmann, 1992).
Education was measured as categorical variables that re-
flect low, middle, and high levels of education for the re-
spondent and for his or her most highly educated child. Be-
cause educational attainment increased substantially over
the course of a generation in Taiwan, the actual years of ed-
ucation reflecting low, middle, and high differ between the
parents and children. For the parents, low education was
considered to be those with no formal schooling, those with
primary education were coded as having middle education,
and those with more than primary had high education. For
children, low education was considered to be those who had
no more than primary. Those with junior or senior high
were coded as having middle education. Those with at least
some college had high education.
Family sizes in Taiwan tend to be large, and each child
has the potential to influence their parent. Ideally, one
would try to bring in the characteristics of each child to the
model, but aside from considering some composite of all
children’s education, this type of modeling is complex and
difficult to interpret. To capture family resources available
through children, we considered the educational attainment
of the most highly educated child. This strategy has at least
two advantages. First, there tends to be a fair amount of ho-
mogeneity in the educational attainment of all children
within a household, so that the most highly educated child is
a good proxy for the education of all children. Second, the
health of an older adult is likely to be most influenced by
the child who has the greatest resources to bring to bear, and
this is well represented by the child with the most education.
We note that rather than indicating the education of a se-
lected child, our measure can be considered a general indi-
cator of resources that may be available to parents through
their offspring.
Because more than one child within a family can have the
same level of education (for instance, more than one child
can have a college education), we could not select a particu-
lar child who has the highest level of education. However,
by determining the children who tie with the highest level of
education, we could construct other variables of relevance.
First, we measured the gender of the most highly educated
child or the gender composition when there were ties. Gen-
Finally, to adjust results for economic well-being and to
assure that influences of education were not due strictly to
wealth, we included a measure of household income. This
comes from the question “At present, how much is the total
monetary income you (and your spouse) receive in a
month?” The survey recorded responses categorically, and
we created an income variable with five categories. Those

WHOSE EDUCATION COUNTS?
S27
Table 2. Variable Descriptions
Coding
Variable
Distribution
0 ꢀ 63.3%, 1 ꢀ 36.7%
1 ꢀ 43.0%, 2 ꢀ 38.1%, 3 ꢀ 18.8%
Low ꢀ 50.7%, Mid ꢀ 30.6%, High ꢀ 18.7%
Existence of a functional limitation
Functional limitation severity
Education of parent
0 ꢀ no, 1 ꢀ yes
1 ꢀ mild, 2 ꢀ moderate, 3 ꢀ severe
Low ꢀ None (comparison category), Mid ꢀ Primary,
High ꢀ More than primary
Education of child
Low ꢀ No more than primary (comparison category),
Mid ꢀ Junior/Senior high, High ꢀ College
Female (comparison category), Male, Both
Coresident (comparison category),
Lives nearby ꢀ Same community or town,
Lives farther away ꢀ Different town or out of country
Measured continuously
0 ꢀ Male, 1 ꢀ Female
0 ꢀ Urban, 1 ꢀ Rural
0 ꢀ Other, 1 ꢀ Married
0 ꢀ Other, 1 ꢀ Mainlander
Low ꢀ 19.1%, Mid ꢀ 43.9%, High ꢀ 37.0%
Sex of child
Residency of child
Female ꢀ 15.2%, Male ꢀ 39.0%, Both ꢀ 45.8%
Coresident ꢀ 61.3%, Lives nearby ꢀ 16.6%,
Lives farther away ꢀ 22.1%
Age
Sex
Residence
Marital status
Ethnicity
M ꢀ 68.3, (SD) ꢀ 6.50
0 ꢀ 55.3%, 1 ꢀ 44.7%
0 ꢀ 64.9%, 1 ꢀ 35.1%
0 ꢀ 32.7%, 1 ꢀ 67.3%
0 ꢀ 81.0%, 1 ꢀ 19.0%
M ꢀ 4.83, (SD) ꢀ 2.15
Lowest ꢀ 26.9%, Second ꢀ 18.3%,
Third ꢀ 25.9%, Highest ꢀ 15.6%,
Missing ꢀ 13.3%
Number of living children
Income
Measured continuously
Lowest (less than 5,000 NT per month),
Second (5,000–9,999 NT), Third (10,000–19,999 NT),
Highest (20,000 NT and over), Missing
Note: NT ꢀ New Taiwan Dollars, the equivalent of about $150 U.S.
in the lowest income category earned less than 5,000 NT
(New Taiwan Dollars, the equivalent of about $150 U.S.)
per month. The next categories are those reporting 5,000 to
9,999 NT, those reporting 10,000 to 19,999 NT, and those
reporting 20,000 NT and higher. Even though responses
were categorical, about 13% of individuals did not report
their income. Rather than imputing a value for these indi-
viduals, we considered those with missing income as a fifth
category, and we examined whether those with missing in-
come had different functioning outcomes than others. Dis-
tributions and coding schemes for all the variables involved
in the analysis are provided in Table 2.
clude gender of child with education of child, proximity of
residence of child with education of child, and education of
respondent with education of child.
For the first part of the analysis we used logistic regres-
sion and assumed that the probability of the existence of a
limitation can be calculated as:
Pr(yj ꢀ 1  Xj) ꢀ [exp (ꢁ1x1j ꢂ ꢁ2x2j ꢂ … ꢁkxkj)] /
[1 ꢂ exp (ꢁ1x1j ꢂ ꢁ2x2j ꢂ … ꢁkxkj)].
Statistical Analysis
The two outcome measures suggest a two-part analytical
approach. According to Manning, Duan, and Rogers (1987)
a two-part strategy is also appropriate when zero values are
considered to be true zeros rather than censored cases. This
is the case in the present problem, as an individual without a
limitation is a true zero with respect to severity and hence is
omitted from the second part of the analysis. The model that
will be used is diagramed in Figure 1. In the first set of
equations we examined the effects of covariates on having a
limitation, using the entire sample. In the second set of
equations, we limited analyses to the approximately 1,400
cases reporting at least one limitation and examined effects
of covariates on limitation severity.
Models were constructed hierarchically. We first exam-
ined the impact of the parent’s education, controlling for pa-
rental characteristics. Next, to view the added influences of
the child, we added child’s education to the model and
hence adjusted parent and child education simultaneously.
We then added the additional covariates that relate to the
children, including their sex and proximity of residence. Fi-
nally, we considered interaction effects that could influence
the way in which child’s education influences the functional
limitation status of their parents. In the figure, dashed lines
indicate these interactions. The interaction effects tested in-
Figure 1. Model to test for the impact of education on functional
status of older adults in Taiwan.

S28
ZIMMER ET AL.
Table 3. Logistic Regression Adjusted Odds Ratios for the
Existence of any Functional Limitations (N ꢀ 3,726)
tional limitation, as did about 340 with midlevel education.
One of the aims of the second stage, outlined above, was to
determine whether these 480 respondents displayed differ-
ent patterns of limitation severity than did those with limita-
tions and no education. In addition, to comment on the ro-
bustness of these findings, we conducted several sensitivity
tests, by first coding functional limitation severity as or-
dered and, as tends to be more traditional, continuous level
variables that include all respondents, and then running ap-
propriate models. These tests are reported in the Results
section below.
Variable
Model 1
Model 2
Model 3
Respondent’s education
Low
1.00
1.00
1.00
Mid
High
.72**
.47**
.77**
.54**
.78**
.58**
Child’s education
Low
Mid
1.00
.88
.69**
1.00
.80*
.61**
High
Child’s sex
Female
Male
1.00
1.35
.87
RESULTS
We begin by examining education’s association with the
existence of any functional limitations. Model 1 of Table 3
shows that, in addition to age, gender, and income, the level
of education of the aged parent, entered separately, is a sig-
nificant determinant of any limitations. This is the standard
finding when considering an individualistic approach to
modeling the functional health of an older adult. For exam-
ple, odds ratios show that older adults who have high-level
education (that is, more than primary), are 53% less likely
to report a functional limitation in comparison to those with
low-level education.
Male and female
Child’s residence status
(comparison ꢀ lives farther away)
Lives farther away
Coresides
1.00
.87
.90
1.08**
1.82**
1.09
.89
1.24
1.00
Lives near by
Age
1.08**
1.75**
1.07
.90
1.21
.97
1.08**
1.81**
1.08
.89
1.25
.98
Sex (1 ꢀ female)
Marital status (1 ꢀ married)
Residency (1 ꢀ rural)
Ethnicity (1 ꢀ mainland)
Number of living children
Income
Model 2 shows that both parent and child effects are sub-
stantial and significant predictors of having a functional
limitation, adjusting for other sociodemographic covariates.
Comparing Model 1 with Model 2, we found the effect of
parent’s education is reduced somewhat. The odds ratios
change from .72 to .77 when the older adult has midlevel
education in comparison with low level, and from .47 to .54
when the older adult has high-level education in comparison
with low level. Net of parent’s education, however, the child’s
education appears to have an additional effect. For example,
the odds ratio for having a limitation for those with high
versus low child education is .69, meaning that the those
who have a child with high education are 31% less likely to
report a limitation in comparison with those whose most
highly educated child has primary education or less. Log-
likelihood results indicate that each education measure, that
of child and their aged parent, are significant factors in de-
termining the probability of reporting a functional limitation
net of each other.
Model 3, examining whether residence proximity or gen-
der of child makes a difference with respect to functional
limitations of the aged parent, shows little improvement in
log-likelihood, and the added coefficients are not signifi-
cant. This indicates that a child living farther away versus
coresiding or living nearby, and the most highly educated
child being a woman versus a man versus being either male
or female, does not matter with respect to having a func-
tional limitation. Adding these variables does make the con-
trast between midlevel child schooling and no schooling
significant.
Lowest
Second
Third
Highest
Missing
ꢃ2 Log-Likelihood
1.00
1.00
1.00
.70**
.52**
.42**
1.21
.73**
.55**
.46**
1.23
.73**
.55**
.46**
1.23
4,241.0
4,229.7
4,222.5
*p ꢄ .05; **p ꢄ .01.
In the second part, we used a cumulative ordered logistic re-
gression approach (Agresti, 1996), which has the following
form for predicting the probability of membership in any
group:
Pr(outcomej ꢀ i) ꢀ Pr(kj ꢃ 1 ꢄ ꢁ1x1j ꢂ ꢁ2x2j ꢂ … ꢁkxkj ꢂ
ꢅj ꢆ ki)
where k is the cut point between any two groups and x1,x2 . . .
xk are the variables listed on the left-hand side of Figure 1.
The ordinal logistic model is a proportional model in that it
assesses the log odds ratio changes to be the same when di-
viding the ordered variable into any two parts. For instance,
in a three-category ordered variable, coefficients represent
the change that a one-unit increase in an independent vari-
able has on the log odds of membership in Group 1 versus 2
or 3, or Group 1 or 2 versus 3. Probabilities were predicted
on the basis of one set of coefficients and a number of con-
stants, usually referred to as cut points, for which the num-
ber of constants is the number of categories in the ordered
variable minus one.
A disadvantage of the two-part strategy for modeling
having any limitations and severity of limitations is that the
variation in older adult education, already lesser in Taiwan
than is the case in most Western-based samples, narrows
when eliminating those who have no limitations from the
second stage of the analysis. But, some variation remained.
About 140 respondents who had high education had a func-
Table 4 presents odds ratios from ordinal logistic regres-
sion models that predict the functional limitation severity
for older adults who reported at least one functional task
difficulty. Looking first at Model 1, we found the education
of the older adult has some negative association with limita-
tion severity, but the coefficients are not significant. This is

WHOSE EDUCATION COUNTS?
S29
Table 4. Ordered Logistic Regression Adjusted Odds Ratios for
Functional Limitation Severity for Those Reporting at Least One
Functional Limitation (N ꢀ 1,369)
cated child is male, female, or either. Child’s education in-
fluences functional status consistently across these other co-
variates.
When testing for a series of interaction effects between
child’s education and other characteristics, we found there
was virtually no change to our conclusions and the interac-
tions themselves were generally insignificant. For this rea-
son, we do not report these results in tabular form (they are
available from the authors on request). There is no interac-
tion between child and parent education, suggesting that the
impact of child’s education is consistent regardless of the ed-
ucation of the parent. One interaction worth noting is the
impact of the interaction of child’s low education with liv-
ing farther away from parents on severity of limitations. Se-
verity is significantly greater in cases in which the child has
low education and lives far away from their parents versus
any other combination of education and proximity.
Variable
Model 1
Model 2
Model 3
Respondent’s education
Low
Mid
High
Child’s education
Low
Mid
High
Child’s sex
Female
1.00
.77
.99
1.00
.80
1.07
1.00
.82
1.13
1.00
.78
.69*
1.00
.68**
.59**
1.00
1.10
.87
Male
Male and female
Child’s residence status
Lives farther away
Coresides
1.00
.90
To determine the robustness of our findings with respect
to severity, we conducted several sensitivity analyses in-
volving modeling severity of limitations without eliminat-
ing the group reporting no limitations. In one test we coded
severity as an interval variable, summing scores of 0 for
having no problem, 1 for a little difficulty, 2 for a lot of dif-
ficulty, and 3 for being unable to do the task across the five
functional tasks. This is a more traditional way of coding
limitation severity. We then ran least squares multiple re-
gression to predict this score, which ranged from 0 to a
maximum of 15. This procedure confirmed the robustness
of our findings and therefore adds confidence to the results
presented above. In the multiple regression, child’s educa-
tion was both a stronger determinant and more significant
than respondent’s education. (For instance, the unstandard-
ized beta comparing respondents with low to high education
is .297, p ꢀ .064, whereas the coefficient comparing chil-
dren with low to high education is .858, p ꢀ .000). In fur-
ther tests we ran an ordered regression coding having no
limitations as an additional category and a multinomial re-
gression, treating limitation severity as a discretely mea-
sured variable, including a category for no limitations. Al-
though all of these tests do not disentangle the effects we
are modeling—the presence of a limitation versus the sever-
ity of limitations—they are all in agreement that the child’s
education is strongly associated with the limitation mea-
sures used, and that at times it is a statistically significant
predictor, whereas the education of the older adult is not.
These tests are not reported here in tabular form, but are
available from the authors on request.
Lives near by
Age
1.01
1.05**
1.05
1.17
1.12
1.18
1.08**
1.05**
1.04
1.14
1.15
1.08
1.05**
1.05
1.16
1.13
1.15
Sex (1 ꢀ female)
Marital status (1 ꢀ married)
Residency (1 ꢀ rural)
Ethnicity (1 ꢀ mainland)
Number of living children
Income
Lowest
Second
Third
Highest
1.05*
1.06*
1.00
.91
.59**
.78
1.40*
2,741.8
1.00
.70
.62**
.83
1.43*
2,709.4
1.00
.96
.62**
.90
1.42*
2,704.6
Missing
ꢃ2 Log-Likelihood
*p ꢄ .05; **p ꢄ .01.
a result that is somewhat surprising and different than when
considering the existence of any limitation. Model 2 indi-
cates that the education of the most highly educated adult
child does indeed have a significant impact on severity of
limitations over and above the effect of parent’s education.
The odds ratios for midlevel and high education are less
than 1.00, indicating that in cases in which the education of
the child is higher, the probability that limitations are more
severe decreases, although only the coefficient for high edu-
cation is statistically significant. This odds ratio suggests
that when the highest level of education among children is
high, the probability that the parent will have a higher level
of limitation severity decreases by about 31% in compari-
son with a situation in which the highest level of education
among children is low. This result means that only the addi-
tion of the education of the child significantly enhances the
predictive power of the model and that child’s education
adds something net of the older adult’s education, but par-
ent’s education does not add to the model net of their other
sociodemographic characteristics.
DISCUSSION
The current study examined the impact of education on
functional status outcomes of older adults in Taiwan. Its
contribution comes in the attempt to ascertain whether the
level of education of an adult child has added effects over
and above an older adult’s own education and in the dual
conceptualization of functional limitations as existence and
severity. We chose Taiwan as an ideal environment to test
this phenomenon because it is characterized by high degrees
of family integration despite rapid changes in socioeco-
nomic structure. It is also characterized by wide discrepan-
The addition of residency status of the child and gender
of the child show similar results in predicting limitation se-
verity as they did in predicting any limitations. That is,
these effects are not significant themselves but enhance the
effect of children’s education. It does not seem to matter
then where the child lives or whether the most highly edu-

S30
ZIMMER ET AL.
cies in parent–child education levels, suggesting that an
adult child may have the ability to mobilize additional re-
sources that can be brought to bear on the parent’s behalf.
The evidence overwhelmingly supports the hypothesis that
the educational level of children can be an added influence
on their parent’s functional status. Child’s education proved
to be important in two ways. First, it had an important addi-
tive influence on the existence of a functional limitation.
Second, when examining severity, child’s education was
shown to be the crucial determinant. Conversely, the educa-
tion of an older adult was a significant determinant in pre-
dicting the existence of functional limitations—a finding
that has been found consistently in the literature—but not
the severity of the limitations. This distinction has gone un-
observed in previous research, likely because existence and
severity have not generally been isolated in predictive mod-
els. Because a majority of older adults have no limitations,
results from any conceptualization of functional status that
does not separate existence and severity would be heavily
weighted toward no limitations versus any limitations,
whereas the specific determinants of the level of severity,
among those with any limitations, may go unnoticed.
As noted, the significance of an older adult’s education on
the existence of a functional limitation confirms much previ-
ous research, this time in Taiwan. The results for the severity
of limitations shed some further light on the determinants of
functional status among this sample of older adults and allow
us to speculate further on the dynamics involved. The func-
tion of own level of education on late life functional status in
Taiwan may be mainly one of prevention. Education is deter-
mined early on in life, and for older adults, the influence of
education is a life-long phenomenon. Lifestyle factors could
influence individuals from an early age. These include health
behaviors, such as eating a healthy diet or partaking in activi-
ties and exercise. Next, there is a broad range of psychosocial
phenomena, including stress, an important determinant of
health (Pearlin, 1989); locus of control, or a sense of control
over the course of one’s life; environmental factors, some of
which may be related to occupation; and social support
(House et al., 1994). There also are resources that become
available and can be used throughout life that allow for ac-
cess and use of health services. Those with higher education
can better afford quality and more constant medical care
throughout life. These influences can serve to ward off poten-
tial health problems that relate to physical functioning in later
life.
ranging care, the financial resources that they can mobilize,
and their advice on how to deal with a given ailment can de-
termine the course of the disorder. A possible specific
mechanism associated with children’s resources is that
more highly educated children are more likely to be em-
ployed in situations in which they have health insurance,
and it was not uncommon in Taiwan (prior to the 1995 en-
actment of the Universal Health Plan) to allow parents as
well as children to be listed as dependents, making them eli-
gible for coverage.
Several competing hypotheses to that proposed here must
be considered. One is that the lack of significance in par-
ent’s education on severity is a statistical artifact due to the
reduction in the variation in education that occurs when
eliminating a good proportion of the more highly educated
people who have no limitations in the second part of the
analysis. In response we note that although there is a strong
association between the educational level of the respondents
and the existence of a limitation, there was a sufficient num-
ber of cases of more highly educated parents with a limita-
tion to support the two-stage strategy. In addition, the sensi-
tivity tests we conducted that did not use the two-step
strategy reaffirm the importance of children’s education vis
a vis the parents’. Indeed, our findings can be interpreted as
establishing the importance of tracing the health implica-
tions of the strong education generation gap and the rapid
widening of educational opportunities currently in place in
Taiwan. Although the educational level of children is corre-
lated with the educational level of their parents in Taiwan as
elsewhere, the changing opportunity structure means that
more than 20% of older adults with no education have chil-
dren who are college educated. It is reasonable to expect
that the resources and knowledge possessed by these chil-
dren would benefit the health outcomes of their parents as
well as other aspects of their parents’ lives.
Another competing possibility is the existence of an en-
dogenous association between the goals of educating a child
and the resources, attitudes, and practices of a parent that
also promote his or her good health. It may be that parents
who have been attentive to maintaining their own health
through use of available health care facilities and knowl-
edge would also be those most likely to be attentive to the
rapidly widening educational opportunities in Taiwan and
to be most successful in furthering their children’s educa-
tional attainment. It is also possible that the mechanism hy-
pothesized above—of the mustering of resources at the time
of limitation—is misplaced in time. In a society like Tai-
wan, with frequent contacts between parents and children,
the high level of education among children may redound to
the advantage of parents’ health throughout their later life.
If so, it does not negate the benefits of the child’s education
but suggests additional mechanisms at work. The fact that
proximity of the children with higher education was not it-
self significant in the analysis runs counter to this possibil-
ity, but more testing along these lines is in order.
The associations with severity substantiate the role of
family in the progression of limitations once they arise. In
Chinese societies, in which filial piety is a moral obligation,
the health care of an older adult is the responsibility of the
family. Once a functional limitation occurs, the resources,
knowledge, and assistance that can be provided by adult
children come into play, and it appears from our results that
the type of resources that become mobilized are captured
well by the education of the most highly educated child.
Those with more highly educated children gain similarly
with respect to the level of severity regardless of their own
education, and the gain is related to the level of the chil-
dren’s education. This suggests that, when a parent is ailing,
the quality of the child’s assistance in caregiving and ar-
Despite these cautions, the hypothesis advanced here is
that the quality of children’s resources, as reflected in their
education, that can be mobilized at the time of parental func-
tional limitation is influential in the progression of that limita-
tion. Our findings that children’s education is more strongly

WHOSE EDUCATION COUNTS?
S31
Address correspondence to Zachary Zimmer, The Population Council,
One Dag Hammerskjold Plaza, New York, NY 10017. E-mail: zzimmer@
popcouncil.org
associated with the severity of the limitation than is its pres-
ence and that parental education ceases to be significantly re-
lated to severity point strongly in this direction. Nevertheless,
much more testing remains to be done in terms of replication
as well as in terms of more detailed data collection. There are
opportunities for replication in several directions. A number
of developing and newly industrialized countries are similar
to Taiwan in the rapid broadening of educational opportuni-
ties in recent decades, making for substantial differences be-
tween the education of the current older population and their
children, as well as minimizing the level of educational corre-
lation between parents and children. Many of these societies
also have strong norms for support of parents by children and
high levels of coresidence, though there is also a fair degree
of variation on these dimensions. It would be of considerable
interest to replicate the type of analysis presented here across
a set of these countries to observe how consistently and
strongly the role of children’s education emerges and whether
this association varies with indicators of family norms and
structures. Another replication of interest would be for the
United States and other Western industrial societies where
educational differences between generations will be less, on
average, and where there is some debate about the nature and
changes in family solidarity (Luescher & Pillemer, 1998). It
may be that despite a narrower educational generation gap,
parents in the United States benefit from children’s greater
sophistication in health matters as well as the children’s
knowledge of the health system. Observing the degree to
which children’s education matters with regard to parental
health across different cultural environments can be helpful
for understanding the effects of stratification on older adult
health as well as for understanding the nature and function of
family interactions and solidarity. Including diverse ethnic
groups within the United States that are marked both by cul-
tural heterogeneity and differential intergenerational educa-
tional patterns might be particularly revealing.
In addition to appropriate replications using similar data,
there is the need for more nuanced data collection that fo-
cuses on measuring the degree and form of family mobiliza-
tion that takes place on the onset of a limitation and the rela-
tion of those resources to the health outcomes. This would
most likely require a combination of qualitative and quanti-
tative longitudinal measurements involving both older adults
and key family members. Such research, although challeng-
ing, would greatly elucidate the family caregiving and deci-
sion processes associated with the onset of disorders among
older adults, but it would also allow a more detailed exami-
nation of how the characteristics and resources of family
members interact with those of the older individual to affect
health outcomes. The research presented here suggests that
this aspect of family dynamics merits much more attention.
References
Agresti, A. (1996). An introduction to categorical data analysis. New
York: Wiley.
Antonovsky, A. (1967). Social class, life expectancy, and overall mortality.
Milbank Memorial Fund Quarterly, 45, 31–73.
Asis, M. M. B., Domingo, L., Knodel, J., & Kalyani K. (1995). Living ar-
rangements in four Asian countries: A comparative perspective.
Journal of Cross-Cultural Gerontology, 10, 145–274.
Camacho, T. C., Strawbridge, W. J., Cohen, R. D., & Kaplan, G. A.
(1993). Functional ability in the oldest old: Cumulative impact of risk
factors from the preceding two decades. Journal of Aging and Health,
5, 439–454.
Cornman, J. C., Hermalin, A. I., Roan, C. L., & Chang, M. C. (1996). Val-
ues, accommodations and tensions in Taiwanese families: The
perspectives of adult children and their aging parents. Comparative
study of the elderly in Asia (Research report series No. 96–35). Ann Ar-
bor, MI: Population Studies Center.
Crimmins, E. M., & Saito, Y. (1993). Getting better and getting worse:
Transitions in functional status among older Americans. Journal of Ag-
ing and Health, 5, 3–36.
Duffy, M. E., & MacDonald, E. (1990). Determinants of functional health
of older persons. The Gerontologist, 30, 503–509.
Elo, I. T., & Preston, S. H. (1995). Educational differentials in mortality:
United States, 1979–85. Social Science and Medicine, 42, 47–57.
Freedman, V. A., & Martin, L. G. (1998). Understanding trends in func-
tional limitations among older Americans. American Journal of Public
Health, 88, 1457–1462.
Goldman, N., Korenman, S., & Weinstein R. (1995). Marital status and
health among the elderly. Social Science and Medicine, 40, 1717–1730.
Guralnik, J. M., Land, K. C., Blazer, D., Fillenbaum, G. G., & Branch, L.
G. (1993). Educational status and active life expectancy among older
Blacks and Whites. The New England Journal of Medicine, 329, 110–
116.
Hermalin, A. I., Ofstedal, M. B., & Chang, M. C. (1996). Types of support
for the aged and their providers in Taiwan. In T. K. Hareven (Ed.), Ag-
ing and generational relations over the life course: A historical and
cross-cultural perspective (pp. 400–437). New York: Walter de
Gruyter.
Hermalin, A. I., Ofstedal, M. B., & Lee, M. L. (1992). Characteristics of
children and intergenerational transfers. Comparative study of the el-
derly in Asia (Research report series No. 92–21). Ann Arbor, MI:
Population Studies Center.
House, J. S., Kessler, R. C., Herzog, A. R., Mero, R. P., Kinney, A. M., &
Breslow, M. J. (1990). Age, socioeconomic status, and health. The Mil-
bank Quarterly, 68, 383–411.
House, J. S., Lepkowski, J. M., Kinney, A. M., Mero, R. P., Kessler, R. C.,
& Herzog, R. A. (1994). The social stratification of aging and health.
Journal of Health and Social Behavior, 35, 213–234.
Kaplan, G. A., Strawbridge, W. J., Camacho, T., & Cohen, R. D. (1993).
Factors associated with change in physical functioning in the elderly: A
six-year prospective study. Journal of Aging and Health, 5, 140–153.
Katz, S., Ford, A. B., Moskowitz, R. W., Jackson, B. A., & Jaffe, M. W.
(1963). Studies of illness in the aged: The index of ADL, a standard-
ized measure of biological and psychosocial function. Journal of the
American Medical Association, 185, 914–919.
Kelly-Hayes, M., Jette, A. M., Wolf, P. A., D’Agostino, R. B., & Odell, P.
M. (1992). Functional limitations and disability among elders in the
Framingham Study. American Journal of Public Health, 82, 841–845.
Kitagawa, E. M., & Hauser, P. M. (1973). Differential mortality in the
United States: A study of socioeconomic epidemiology. Cambridge,
MA: Harvard University Press.
Acknowledgments
Lang, O. (1946). Chinese family and society. New Haven, CT: Yale Uni-
versity Press.
Lawton, M. P., & Brody, E. (1969). Assessment of older people: Self-
maintaining and instrumental activities of daily living. The Gerontolo-
gist, 9, 179–186.
This research is supported by The National Institutes of Health–National
Institute of Aging Research Grant 1 R03 AG17270-01.
An earlier version of this article was presented at the Asia-Pacific Re-
gional Conference for the International Year of Older Persons in Hong
Kong, in April 1999.
Link, B. G., & Phelan, J. (1995). Social conditions as fundamental causes
of disease. Journal of Health and Social Behavior, (Suppl.), 80–94.
We thank three anonymous reviewers for their helpful comments.

S32
ZIMMER ET AL.
Luescher, K., & Pillemer, K. (1998). Intergenerational ambivalence: A
new approach to the study of parent–child relations in later life. Jour-
nal of Marriage and the Family, 60, 413–425.
Lynch, J. W., Kaplan, G. A., & Shema, M. S. (1997). Cumulative impact
of sustained economic hardship on physical, cognitive, psychological,
and social functioning. New England Journal of Medicine, 337, 1889–
1895.
Manning, W. G., Duan, N., & Rogers, W. H. (1987). Monte Carlo evidence
on the choice between sample selection and two-part models. Journal
of Econometrics, 35, 59–82.
Marmot, M. G., Rose, G., Shipley, M., & Hamilton, P. J. S. (1978). Em-
ployment grade and coronary heart disease in British civil servants.
Journal of Epidemiology and Community Health, 32, 244–249.
Nagi, S. Z. (1976). An epidemiology of disability among adults in the
United States. Milbank Memorial Fund Quarterly, Health and Society,
54, 439–468.
ily on health: The decade in review. Journal of Marriage and the
Family, 52, 1059–1078.
Ross, C. E., & Wu, C. L. (1995). The links between education and health.
American Sociological Review, 60, 719–745.
Ross, C. E., & Wu, C. L. (1996). Education, age, and the cumulative ad-
vantage in health. Journal of Health and Social Behavior, 37, 104–120.
Silverstein, M., & Bengston, V. L. (1997). Intergenerational solidarity and
the structure of adult child–parent relationships in American families.
American Journal of Sociology, 103, 429–460.
Simonsick, E. M., Kasper, J. D., Guralnik, J. M., Bandeen-Roche, K., Fer-
rucci, L., Hirch, R., Leveille, S., Taina, R., & Fried, L. P. (2001).
Severity of upper and lower extremity functional limitation: Scale de-
velopment and validation with self-report and performance-based
measures of physical function. Journal of Gerontology: Social Sci-
ences, 56B, S10–S19.
Taiwan Provincial Institute of Family Planning. (1989). 1989 survey of
health and living status of the elderly in Taiwan: Questionnaire and
survey design. Comparative Study of the Elderly in Four Asian Coun-
tries (Research Report No. 1). Ann Arbor, MI: Population Studies
Center.
Nagi, S. Z. (1991). Disability concepts revisited: Implications for preven-
tion. In A. M. Pope & A. R. Tarlov (Eds.), Disability in America:
Toward a national agenda for prevention (pp. 309–327). Washington
DC: National Academy Press.
Ortmeyer, C. F. (1974). Variations in mortality morbidity and health care
by marital status. In C. E. Erhardt & J. E. Berlin (Eds.), Mortality and
morbidity in the United States (pp. 159–588). Cambridge, MA: Har-
vard University Press.
Pappas, G., Queen, S., Hadden, W., & Fisher, G. (1993). The increasing
disparity in mortality between socioeconomic groups in the United
States, 1960 and 1986. New England Journal of Medicine, 329, 103–
109.
Pearlin, L. I. (1989). The sociological study of stress. Journal of Health
and Social Behavior, 30, 241–256.
Preston, S. H., & Elo, I. T. (1995). Are educational differentials in adult
mortality increasing in the United States? Journal of Aging and Health,
7, 476–496.
Townsend, P., & Davidson, N. (1982). Social inequalities in health: The
Black report. Harmondsworth, England: Penguin Books.
Verbrugge, L. M. (1989). The twains meet: Empirical explanations for sex
differences in health and mortality. Journal of Health and Social Be-
havior, 30, 282–304.
Verbrugge, L. M., & Jette, A. M. (1994). The disablement process. Social
Science and Medicine, 38, 1–14.
Williams, D. R. (1990). Socioeconomic differentials in health: A review
and redirection. Social Psychology Quarterly, 53, 81–99.
Winkleby, M. A., Jatulis, D. E., Frank, E., & Fortmann, S. P. (1992). So-
cioeconomic status and health: How education, income and occupation
contribute to risk factors for cardiovascular disease. American Journal
of Public Health, 82, 816–820.
Robert, S. A., & Li, L. W. (2001). Age variation in the relationship be-
tween community socioeconomic status and adult health. Research on
Aging, 23, 233–258.
Rogers, R. G., Rogers A., & Belanger, A. (1992). Disability-free life
among the elderly in the United States: Sociodemographic correlates of
functional health. Journal of Aging and Health, 4, 19–42.
Ross, C. E., Mirowsky, J., & Goldsteen, K. (1990). The impact of the fam-
Received January 29, 2001
Accepted June 5, 2001
Decision Editor: Fredric D. Wolinsky, PhD