Full text of DOI:10.1093/geronb/56.4.S206

Journal of Gerontology: SOCIAL SCIENCES
Copyright 2001 by The Gerontological Society of America
2001, Vol. 56B, No. 4, S206–S218
Persistent, Consistent, Widespread, and Robust?
Another Look at Recent Trends in Old-Age Disability
Robert F. Schoeni,1 Vicki A. Freedman,2 and Robert B. Wallace3
1RAND, Santa Monica, California.
2Polisher Research Institute, Philadelphia Geriatric Center, PA.
3College of Public Health, University of Iowa, Iowa City.
Objective. The purpose of this study was to provide new evidence on disability trends among elderly persons from
1982 to 1996.
Methods. The sample includes 124,949 participants aged 70 and older in the 1982–1996 National Health Interview
Surveys. Logistic analysis was used to estimate the trend in disability prevalence after controlling for various sociode-
mographic factors.
Results. We found that: (a) the prevalence of disability has declined, but the gains did not persist throughout the en-
tire period or accelerate over time; (b) only routine care disability has declined, whereas more severe personal care dis-
ability shows no improvements; (c) estimates are robust to the exclusion of the nursing home population but may be
sensitive to growth in the assisted living population; (d) estimates of decline in disability prevalence are fairly consistent
across five national surveys; (e) gains have been concentrated among the most educated elderly persons; and (f) gains in
education appear to be an important confounder of the improvements.
Discussion. Evidence from several surveys using various measures indicates that disability has declined among el-
derly persons. Determining the causes of the improvements should be a high priority in future research efforts.
OR more than two decades, researchers have debated
the implications of prolonged life for the health of older
In fact, much of what is known about trends in disability
prevalence in the 1980s and early 1990s is based upon data
drawn from a limited number of time points. Manton and
colleagues (1993, 1997), for example, base their conclusion
that the decline in the disability prevalence rate has acceler-
ated on just four years of data (1982, 1984, 1989, and 1994)
from the National Long Term Care Survey (NLTCS). Simi-
larly, Waidmann and Liu (2000) conclude that disability
trends have improved based upon only 5 consecutive years
of data (1992 to 1996) from the Medicare Current Benefi-
ciary Survey (MCBS). Freedman and Martin’s (1998) find-
ings of improvements in functioning focus on two of four
available time points over a 9-year period (1984 and 1993)
from the Survey of Income and Program Participation
(SIPP). They reach a similar conclusion (Freedman & Mar-
tin, 2000) based upon the 1984 and 1994 Supplements on
Aging (SOA) to the National Health Interview Survey
(NHIS).
Only one study of this period used more than four time
points to examine disability trends (Crimmins et al., 1997).
Drawing upon annual data from 1982 to 1993 from the
NHIS, the authors examined changes in the prevalence of
personal care disability and changes in the prevalence of
routine care (without personal care) disability. For each type
of disability, the authors plotted age–sex specific prevalence
rates and commented that their visual examination of the
data did not indicate any clear time trend in either disability
measure. They also estimated logistic regression models,
with year entered as a continuous variable, and controls for
shifts in the mean age and sex distribution over time. The
F
Americans (Agree & Freedman, 1999). Indeed, a major fo-
cus of research has been documenting the direction and na-
ture of health and disability trends among older Americans.
Such information is clearly crucial for planning for the fu-
ture health and retirement needs of older Americans.
Studies focused on the 1970s suggested that longer life im-
plied worsening health, as measured by increases in self-
reported disability and chronic disease (Colvez & Blanchet,
1981; Verbrugge, 1984). During the 1980s and early 1990s,
however, disability and functional limitation prevalence rates
among older Americans appeared to decline (Crimmins,
Saito, & Reynolds, 1997; Freedman & Martin, 1998, 2000;
Manton, Corder, & Stallard, 1993, 1997; Waidmann & Liu,
2000), despite what appears to be continued increases in re-
ports of chronic disease (Freedman & Martin, 2000).
Whether such trends in disability are real, driven by im-
provements in the underlying health or social environment
of older Americans, or simply a statistical artifact stemming
from methodological and conceptual problems, is not a new
question (Waidmann, Bound, & Schoenbaum, 1995; Wilson
& Drury, 1984). Potential threats to the validity of compari-
sons of survey responses over time include changes in ques-
tion wording, survey coverage, mode, response rates, proxy
response rates, and—for longitudinal surveys—loss to fol-
low-up (Freedman & Soldo, 1994). Further, long-term un-
derlying trends may not be apparent from—or may be in-
consistent with—evidence relayed by a limited number of
data points.
S206

TRENDS IN OLD-AGE DISABILITY
S207
statistical analysis showed that when age and sex were con-
trolled, there was a significantly lower relative likelihood of
disability in later years whether the two types of disability
were combined or separated. Based on these analyses, the
authors concluded that “although the data indicate some sta-
tistically significant differences between years, there is no
clear ongoing trend in prevalence of disability among older
Americans in the latter part of the 1980s and the early
1990s” (Crimmins et al., 1997, p. S67).
The analysis by Crimmins and colleagues raises several
important questions. First, although these researchers put
their findings in context with those published at the time
(e.g., Manton et al., 1993), several new studies using alter-
native data sources for similar or more recent time periods
are now available. Consequently, revisiting the issues of
whether the NHIS shows declines in disability over a longer
period and whether these results are consistent with those
newer efforts is of interest. Further, although Crimmins and
coworkers stratify their results by age and sex, the analysis
did not investigate whether disability declines were occur-
ring within particular subgroups of the population (e.g., by
race and education level). Moreover, their analysis did not
explore whether findings were robust to a variety of specifi-
cations with respect to time, to the omission of the institu-
tionalized population, or to changes in the socioeconomic
composition of the older population.
With respect to the breadth of disability declines, there is
mixed evidence as to whether recent improvements have
occurred across all demographic and socioeconomic groups.
At least one study using the NLTCS suggests that older Af-
rican Americans did not experience the gains of their White
counterparts with respect to disability (Clark, 1997). In con-
trast, using the SIPP, Freedman and Martin (1998) report
improvements in four areas of functioning—seeing, climb-
ing stairs, lifting and carrying, and walking three blocks—
across almost all demographic and socioeconomic groups,
including African Americans.
Changes in the demographic and socioeconomic compo-
sition of the older population do not appear to account for
recent declines in disability. Even after controlling for shifts
in the age, sex, race, education, and marital status composi-
tion of the older population, Waidmann and Liu (2000) find
statistically significant declines in those needing help with
only instrumental activities of daily living (IADLs), those
needing help with activities of daily living (ADLs), and
those disabled and living in an institution. Similarly, Freed-
man and Martin (1998) find large and statistically signifi-
cant declines in difficulty with four functional limitations
even after controlling for age, sex, race, ethnicity, educa-
tion, marital status, ownership of liquid assets, and region of
residence. These and subsequent analyses (Freedman &
Martin, 1999) suggest that shifts in the educational compo-
sition have strong linkages to recent improvements, al-
though the complex mechanisms underlying this relation-
ship are not well understood.
1990s persisted into the mid-1990s. That is, we address
whether patterns have been persistent over the 15-year pe-
riod. Second, we place findings from the NHIS in context
by comparing our results to published findings from other
national surveys. That is, we attempt to address whether
patterns are consistent across five national surveys of older
Americans. Third, we explore whether declines in personal
care and only routine care disability were experienced
among subgroups of the elderly population. In doing so, we
examine whether patterns are widespread among older
Americans. Finally, we explore whether our findings with
respect to time trends are robust with respect to various
specifications of the time-disability relationship, to the ex-
clusion of the institutional population, and to changes in the
socioeconomic and demographic composition of the older
population that have occurred over this period.
METHODS
Data
The NHIS is a repeated cross-sectional survey of the non-
institutionalized population in the United States. Conducted
annually by the National Center for Health Statistics, the
NHIS includes a sample of roughly 8,000 adults aged 70
and older in each year. These large samples allow relatively
precise estimates of disability among elderly persons for
each year, including estimates for some major subgroups.
The sampling plan follows a multistage area probability de-
sign that permits the representative sampling of households.
The “final basic weights,” which have been poststratified to
represent the civilian noninstitutionalized population, are
used in the calculation of descriptive statistics to adjust for
this design. Although not everyone within a sampled house-
hold is interviewed, information is collected on all house-
hold members by using proxy respondents. The proxies
(i.e., other household members who are in fact interviewed)
report the requested information pertaining to people who
cannot be interviewed themselves.
Disability among people aged 70 and older (71 and older
in 1982) is measured by two questions. The first question
asks about ADL-type limitations: “Because of any impair-
ment or health problem, does ___ need help of other persons
with personal care needs, such as eating, bathing, dressing,
or getting around this home?” Respondents who answered
no to this question were then asked about IADL-type limita-
tions: “Because of any impairment or health problem, does
___ need help of other persons in handling routine needs,
such as everyday household chores, doing necessary busi-
ness, shopping, or getting around for other purposes?” Be-
cause the questions were changed substantially in 1982 and
again in 1997, we examined the period 1982–1996. The re-
sulting sample includes 124,949 men and women aged 70
and older over the 1982–1996 period.
For the goal of examining trends in disability, the NHIS
has several advantages over other data sets. Perhaps most
important, it is an annual survey administered over a large
number of years. The other national data sets that have been
used to examine trends have fewer data points. In addition,
the NHIS has large, nationally representative samples, a
consistently measured indicator of disability, and it is
This study reassesses the documented decline in disabil-
ity using 15 consecutive years of data from 1982 to 1996
from the NHIS. The research extends the existing literature
on disability trends in several ways. First, we examine
whether the patterns observed in the late 1980s and early

S208
SCHOENI ET AL.
widely used and well documented. Because the NHIS is not
a longitudinal survey, there is no loss to follow-up or learn-
ing effects (i.e., respondents in subsequent waves of a longi-
tudinal survey may begin to answer “no” because they real-
ize the interview will be shorter if they report not needing
help) that may bias estimates of disability prevalence.
The two most important limitations of the NHIS are its
relatively narrow measures of disability and its restriction to
the noninstitutionalized population. Because information
was not available about specific activity limitations or
whether devices were used to help complete various activi-
ties, we were unable to explore whether improvements had
occurred for activities such as bathing and dressing or shop-
ping and housework.
More important, however, because changes in the size and
composition of the elderly institutionalized population can af-
fect the observed trends in disability, we cannot conclude
with certainty that the trends in the NHIS characterize those
of the total resident population aged 70 and older. Specifi-
cally, the NHIS excludes people living in institutionalized
group quarters or members of the military. (The NHIS does
not exclude people in noninstitutionalized group quarters.)
As stated in the technical documentation, “institutionalized
group quarters house people who, in most cases, stay invol-
untarily and are not allowed to come and go without receiv-
ing permission” (U.S. Bureau of the Census, 1997, p. 83). We
interpret this definition to include residents of a nursing
home, because in most cases they cannot leave the facility
without the approval of a medical adviser. It is less clear
whether residents of an assisted living facility would be cap-
tured by this definition; some facilities may place restrictions
on residents’ movement in and out of the building. Or, it may
appear in drawing up the sampling frame that a restriction is
in place, even though it is not. Therefore, we will present
three different estimates of disability prevalence: (a) one that
does not adjust for any component of the institutionalized
population, (b) one that factors in the nursing home popula-
tion, and (c) one that factors in both the nursing home and the
assisted living populations.
There is evidence that the size of the nursing home popu-
lation declined during the 1980s and early 1990s (Bishop,
1999; Manton et al., 1993; Spector, Fleishman, Pezzin, &
Spillman, 1998), but elderly persons living in institutions
are more likely to be disabled (Rhoades & Krause, 1999).
The assisted living population was also substantial by the
mid-1990s, exceeding half a million persons according to a
recent survey (Hawes, Rose, & Phillips, 1999).
To evaluate the importance of omitting the institutional
population, we drew upon information from a variety of
sources. Because data on the institutional population are
available for only a limited number of years, we restricted
this sensitivity analysis to 1985 and 1995. We obtained esti-
mates of the nursing home population aged 70 and older
from the 1985 and 1995 National Nursing Home Surveys
(NNHS): 1,265,365 in 1985 and 1,324,500 in 1995. We de-
termined the number of nursing home residents with IADL-
only disability in each year using the NNHS disability esti-
mates. Specifically, 7.6% of nursing home residents in 1985
had no ADL disabilities (Hing, Sekscenski, & Strahan,
1989). We assume that all residents who did not have an
ADL disability had an IADL disability; therefore, the
IADL-only disability prevalence in the nursing home popu-
lation was 7.6% in 1985. Similarly, the estimate for 1995
was 3.1% (Dey, 1997).
We obtained an indirect estimate of the assisted living
population by subtracting the sum of the NHIS and NNHS
estimates for the 70 and older population from estimates
from the Census Bureau of the total resident population for
1985 and 1995 (U.S. Bureau of the Census, 2000). Using
this indirect method, we found that the number of elderly
persons living in assisted living facilities was small in
1985—53,706. But the size of this population rose substan-
tially by 1995 to 679,500. The most comprehensive, exter-
nal estimate of the number of persons in assisted living fa-
cilities (with 10 beds or more) is 521,500 in 1998 (Hawes et
al., 1999), which is roughly comparable to our indirect esti-
mate.
Finally, we drew our estimate of the prevalence of dis-
ability within the 70 and older assisted living community
(55%) from the 1993 Asset and Health Dynamics Among
the Oldest Old (AHEAD) study (a nationally representative
survey of the community-dwelling population aged 70 and
older that does include elderly persons living in assisted liv-
ing facilities). We assume this estimate holds in both 1985
and 1995, and also explore possible changes in this rate over
time.
Statistical Methods
A logistic model is estimated to determine whether im-
provements in disability are robust to changes in the demo-
graphic composition of the sample—age and sex—as well
as the extent of proxy reporting. The dependent variable is
an indicator that takes the value of 1 if the person has either
an ADL or IADL disability, 0 otherwise. The basic model
includes an indicator for whether the person is female, the
person’s age, and indicators for whether the interview was
completed by a proxy respondent. The models include a lin-
ear age specification. Alternative specifications (quadratic,
cubic, 5-year age categories) led to similar estimated trends;
therefore, for the sake of clarity, we have reported the linear
age specification. The time trend is represented by a simple
linear variable that takes the value 0 in 1982, 1 in 1983, 2 in
1984, . . . and the value 14 in 1996.
Extensions of this model allowed us to test three addi-
tional hypotheses. First, we determined whether the im-
provements have been experienced broadly across the popu-
lation. Interacting the time trend with an indicator for each
group allowed us to test this hypothesis. Differences were
estimated by sex, race, age, marital status, region of resi-
dence, and education. In each case, the direct effect of the
variable under examination (sex, race, age, etc.) was also in-
cluded in the model.
Second, we determined whether the improvements in dis-
ability are confounded by changes in the composition of the
population. For example, more highly educated people are
less likely to be disabled, and the elderly population has be-
come more highly educated over the sample period. There-
fore, this change in educational composition of the population
may account for some of the observed trend in disability.
The NHIS is somewhat limited in its ability to identify

TRENDS IN OLD-AGE DISABILITY
S209
Figure 1. Prevalence of disability among population aged 70 and older: 1982–1996.
proximate determinants of disability, so we were limited to
examining education, marital status, and race. Specifically,
we added these three factors to the logistic model and deter-
mined the extent to which the estimated time trend changed.
Third, we examined both the persistence and robustness
of the improvements by estimating the time trends over var-
ious subperiods of 1982 to 1996. The linear assumption of
the time trend was also relaxed by allowing for a fully non-
parametric specification of the changes across years. Fi-
nally, it was determined whether both the ADL and IADL-
type measures of disability changed in similar ways.
70 is calculated from the 1990 NHIS. Similarly, the disabil-
ity prevalences for the 1920 birth cohort at ages 71, 72, 73,
74, 75, and 76 are estimated from the 1991, 1992, 1993,
1994, 1995, and 1996 NHIS samples, respectively. Linking
these estimates together, we can depict disability age pro-
files separately for several birth cohorts.
Figure 2 demonstrates that there has been some improve-
ment in disability across birth cohorts. Disability prevalence
among 75-year-olds was lower for people born in 1920 than
it was for people born in 1915, and the prevalence for peo-
ple born in 1915 was lower than it was for those born in
1910. For most ages and birth cohorts, age-specific disabil-
ity prevalence is lower for more recent birth cohorts.
RESULTS
Table 1 reports disability prevalence separately for men
and women in each year. The 95% confidence interval (CI)
is reported for each estimate based on the standard error cal-
culation appropriate for these binomial variables (i.e., stan-
dard error equals, p(1 – p) ⁄ n, where p is the mean proba-
Trends
Unadjusted trends.—Figure 1 reports the prevalence of
disability among the noninstitutionalized population aged
70 and older in each year 1982–1996. Roughly 20% of this
population reported needing help with personal or routine
care. Moreover, consistent with Crimmins and colleagues
(1997), there was a decline in disability during this period.
In 1982, 22.7% were disabled, declining to 19.3% in 1996.
This 3.4 percentage point improvement represents, on an
annual basis, a 1.1% decline in disability (i.e., [(3.4/22.7)/
(1996–1982)] ꢀ 100 ꢁ 1.069).
Figure 1 also demonstrates that the decline was concen-
trated within a short period, 1982 to 1986; disability preva-
lence fell by 17% over these 5 years. Despite fluctuations in
subsequent years, by 1996 the prevalence of disability was
virtually the same as it was 10 years earlier. This pattern is
examined further in the multivariate analyses below.
An alternative way to display the changes is by following
birth cohorts through the NHIS, and these estimates are de-
picted in Figure 2. For example, people born in 1920 were
70 years old in 1990, and their disability prevalence at age
Figure 2. Age-specific disability profiles by birth cohort.

S210
SCHOENI ET AL.
bility and n is the number of observations). In addition,
disability is disaggregated into the two types measured in
the NHIS–any ADL disability and IADL disability only.
The table demonstrates that the trends reported in Figure 1
for the entire population hold true for both men and women.
In addition, the table demonstrates that most of the decline
in disability is due to the improvement in having only IADL
disability and not ADL disability. Unlike the estimates of
ADL disability, IADL-only disability experienced a sus-
tained improvement over the entire period, declining from
14.5% in 1982 to 10.9% in 1996.
One potential explanation for changes in disability preva-
lence is that mortality may have changed over the period.
For example, if improvements in mortality result in the sur-
vival of a greater number of frail elderly people, then dis-
ability prevalence may increase. However, the evidence
over the survey period does not appear consistent with this
hypothesis. Figure 3 reports the trend in disability preva-
lence and the trend in life expectancy at age 75 (National
Center for Health Statistics, 1999, Table 28), and these two
time series do not move together very closely. The simple
correlation between 1985 and 1996 is just .07 (data for
1982–1984 were not available for life expectancy).
Adjusting for the institutionalized population.—Table 2
demonstrates the sensitivity of our trend results to the exclu-
sion of the institutionalized population. Row A reports the
disability prevalence rates for the NHIS (i.e., noninstitution-
alized) population; these are a replication of the estimates in
Table 1. Estimates in row B adjust for the nursing home popu-
lation by adding estimates from the NNHS for 1985 and
1995 to the numerator and the denominator of the rate in
row A. Finally, estimates in row C adjust for both the nurs-
ing home population and the assisted living population.
Among the noninstitutionalized population, disability de-
clined from 20.4% to 18.5% over the 10-year period. Once
we take into account the nursing home population, disabil-
ity prevalence is obviously somewhat higher, rising from
20.4% to 25.7% in 1985. However, the improvement over
time is not sensitive to the inclusion of the nursing home
population. Without this population, disability falls by 1.9
percentage points (or 9.3%) from 1985 to 1995; with this
population included, disability falls by 2.5 percentage points
(or 9.8%) over the same period. Moreover, this insensitivity
of the estimated change holds for both the ADL and IADL-
only measures of disability. When both the nursing home
and assisted living populations are taken into account, we
calculate that disability prevalence for the total resident
population aged 70 and older—including elderly persons
living in nursing homes and assisted living facilities—was
25.8% in 1985 (row C in Table 2). Furthermore, we con-
tinue to find that disability improved, but the gains are
somewhat diminished. Instead of a 10-year decline of 2.5
percentage points (9.8%), we find a decline of 1.7 percent-
age points (or 6.5%).
In sum, a large number of disabled elderly persons lived
in assisted living facilities in 1995, whereas very few lived
in these facilities in 1985. Because these facilities may be
inadvertently omitted from the NHIS sampling frame, dis-
ability prevalence among the noninstitutionalized popula-

TRENDS IN OLD-AGE DISABILITY
S211
Figure 3. Disability prevalence and life expectancy at age 75.
tion as estimated in the NHIS may be a downwardly biased
estimate of the disability prevalence within the total resident
population. Moreover, the growth in assisted living facili-
ties has implied that improvements in disability estimated in
the NHIS are somewhat overstated.
One limitation of these analyses is lack of information on
disability prevalence of people in assisted living facilities in
1985. It may be that this population was highly disabled in
1985, but relatively less disabled by 1995. However, the
number of people living in such facilities in 1985 was so
small that this assumption is of no practical importance.
Even if we assumed that all 53,706 residents of assisted liv-
ing facilities in 1985 (row 9 in Table 2) were disabled, dis-
ability prevalence would have fallen by 1.8 percentage
points (or 7.0%).
Comparison with other data sets.—Table 3 reports dis-
ability prevalence estimates from five national surveys that
Table 2. Disability Prevalence Accounting for Institutionalized Population Aged 70 and Older, 1985 and 1995
1985
1995
Change
National Health Interview Survey (NHIS)
1. Total
2. Number disabled
17,753,893
3,627,179
1,225,856
2,401,323
21,689,000
4,016,520
1,660,747
2,355,773
3,935,107
389,341
434,891
ꢂ45,550
3. Number with ADL disability
4. Number with IADL-only disability
National Nursing Home Survey (NNHS)
5. Total
6. Number with ADLa disability
7. Number with IADL-onlya disability
8. Total resident population, U.S. Bureau of the Census
9. Total resident population minus NHIS and NNHS populations (8ꢂ1ꢂ5)
1,265,375
1,169,207
96,169
19,072,972
53,706
1,324,500
1,283,441
41,060
23,693,000
679,500
59,125
114,234
ꢂ55,109
4,620,026
625,794
A. NHIS disability prevalence (as reported in Table 1)
Any disability (2/1)
ADL (3/1)
20.43
6.90
18.52
7.66
ꢂ1.91
0.75
IADL-only (4/1)
13.53
10.86
ꢂ2.66
B. Add to the numerator and denominator, the NNHS population
Any disability (2ꢃ6ꢃ7)/(1ꢃ6ꢃ7)
ADLa (3ꢃ6)/(1ꢃ6ꢃ7)
25.72
12.59
13.13
23.21
12.79
10.41
ꢂ2.52
0.20
ꢂ2.72
IADL-onlya (4ꢃ7)/(1ꢃ6ꢃ7)
C. Incorporate population not captured by NHIS or NNHSb
Any disability [2ꢃ5ꢃ0.55*(9)]/(8)
25.81
24.12
ꢂ1.69
Note: ADL ꢁ activity of daily living; IADL ꢁ instrumental activity of daily living.
aEstimate of the share of the nursing home population with IADL-only disability (7.6% in 1985 and 3.1% in 1995) is drawn from the NNHS in each year.
bAssumes that 55% of elders in assisted living facilities report at least one ADL or IADL in 1985 and 1995; these estimates are drawn from Wave I of AHEAD (As-
set and Health Dynamics Among the Oldest Old).

S212
SCHOENI ET AL.
Table 3. Comparison of Trends in Disability Across Five National Surveys
SIPPd
SOAe
Year
NHISa
NLTCSb
23.7
MCBSc
Seeing
Lifting
Climbing
Walking
Upper Body
Lower Body
34.2
LSOAf
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
22.7
21.8
20.9
20.4
18.9
19.6
19.8
19.5
19.1
20.1
20.7
20.2
19.6
18.5
19.3
23.7
22.6
21.3
21.7
26.6
24.7
25.8
5.1
18.8
21.6
21.3
20.5
17.0
15.7
24.1
24.8
24.6
25.2
24.5
24.5
35.3
34.6
35.3
34.1
32.6
16.8
24.5
25.6
25.3
4.3
28.5
Notes: NLTCS (National Long Term Care Survey) and MCBS (Medicare Current Beneficiary Survey) include institutional population. NLTCS, MCBS, and SIPP
(Survey of Income and Program Participation) are for 65+; NHIS (National Health Interview Survey) and SOA (Supplements on Aging) are for 70+.
aNHIS: disability as defined in text.
bNLTCS: unable to perform ADL or IADL without help for 3 or more months, or in institution (Manton et al., 1997).
cMCBS: any difficulty or receiving help, supervision or using equipment with ADL or IADL or disabled in an institution (Waidmann & Liu, 2000).
dSIPP: any difficulty with task (Freedman & Martin, 1999).
eSOA: unable to carry out upper/lower body tasks (Freedman & Martin, 2000).
fLSOA (Longitudinal Survey on Aging): unable to perform at least one ADL or IADL (Crimmins et al., 1997).
have been used to estimate national trends: the NHIS,
NLTCS, MCBS, SIPP, and SOAs. One additional survey,
the Longitudinal Survey on Aging (LSOA), has also been
used to examine aggregate changes in disability prevalence
(see Crimmins et al., 1997), but we exclude it here because
of several design features that threaten the validity of such
comparisons (e.g., no replenishment of the sample, large
losses to follow-up over time, a shift toward proxy inter-
views, and a change in mode; see Freedman and Soldo,
1994, for further discussion of this point).
The measures of disability vary across the surveys. For
example, the NHIS measures need for help with routine or
personal care needs whereas the NLTCS measures chronic
disability as the need for help for at least 3 months with one
or more ADL or IADL activity or in an institution. Each
survey measure is listed in the table notes.
Estimates from all surveys except the NHIS are available
for only a few years. Four of the six data sets measure dis-
ability in 1984 and 1989/1990, and in three of the surveys
(NHIS, NLTCS, and SIPP) disability declined. The SOA
also shows a decline between the two available years, 1984
and 1995. Three data sets provide estimates in 1989/1990
and 1994. In two of these surveys (NHIS and SIPP) disabil-
ity did not decline, the exception being the NLTCS. At the
same time, the two surveys that provide the most recent esti-
mates of disability (MCBS and NHIS) both find improve-
ments between 1992 and 1996.
The evidence from the various surveys—using somewhat
different measures of disability—is fairly consistent. First,
there was a substantial decline between 1984 and 1990
(NHIS, SIPP, and NLTCS). However, the NHIS, which in-
cludes data on each year during the period, shows that all of
the improvements in the 1980s took place by 1986. Second,
there was an additional decline between 1992 and 1996. The
MCBS and the NHIS show improvements in disability of
8.5% and 6.8% over this 5-year period, respectively. Third,
there was no improvement between 1990 and 1993/1994
(NHIS and SIPP), and there is evidence from the NHIS that
there was no change over the entire period 1986 to 1994.
One data point that is inconsistent with these conclusions is
drawn from the 1994 NLTCS. The NLTCS shows a decline
in disability between 1989 and 1994, which is not found in
the NHIS or the SIPP (for 1990 to 1993).
Logistic Regressions
Have the improvements been experienced broadly?—
Estimates from Model 1 in Table 4 verify the unadjusted
trends. After controlling for age, sex, and proxy reporting,
the time trend implies that with each year, disability de-
clines by 1.2% (i.e., [1.0ꢂ0.988] ꢀ 100). This estimate is
almost identical to the 1.1% decline derived by taking the
change in the unadjusted disability prevalence from 1982 to
1996, as reported in Table 1.
The subsequent columns of Table 4 allow the time trend
to vary across sociodemographic groups. At the bottom of
each column is reported the results of a test for equality of
the trend across the various groups.
The declines were fairly widespread across the popula-
tion. The decline of 1.2% per year was experienced simi-
larly by men and women, which is consistent with findings
reported by Crimmins and colleagues (1997). Whites expe-
rienced a somewhat larger decline than non-Whites (1.6%
vs 1.3%), but the difference was not statistically significant
at standard levels.
Substantial and statistically significant differences were
found by age, marital status, region, and education. The im-
provements were estimated at 1.5% for 70- to 79-year-olds

TRENDS IN OLD-AGE DISABILITY
S213

S214
SCHOENI ET AL.
Table 5. Sample Composition in 1982 and 1996
1982
population changed over this period. Most importantly, edu-
cational attainment improved. In 1982, 17.0% of older
Americans had more than a high school degree, whereas by
1996 this share had increased to 27.0%. By 1996, the older
population had also become slightly younger, more likely to
be married, and less likely to be White.
1996
Education
0–8 years
9–11 years
12 years
0.46
0.16
0.21
0.17
0.22
0.16
0.35
0.27
Estimates in Table 6 examine the factors that confound the
trend by including each factor in the logistic models and de-
termining the extent to which the time trend was weakened.
The bottom row of the table reports the percentage change in
the time trend when these additional variables are included,
relative to the time trend with the baseline specification. For
example, the trend before the control is added was
1.0000ꢂ0.9880 ꢁ 0.0120, or 1.2% per year. The trend after
controlling for education was 1.0000ꢂ0.9958 ꢁ 0.0042, or
0.42% per year, so the change in the trend was 1.2ꢂ0.42 ꢁ
0.78. Therefore, the percent change in the trend was (0.78/
1.2) ꢀ 100 ꢁ 65%. In sum, the change in the educational dis-
tribution was the most important factor confounding the time
trend. Adjusting for marital status also reduced the time trend,
but by only 7%. Adjusting for race actually increased the time
trend. This effect occurred because Whites were less likely to
be disabled, and the share of elders who were White has de-
clined slightly over time (Table 5).
13+ years
Age
70–79 years
80–89 years
90+ years
0.68
0.28
0.03
0.47
0.62
0.09
0.23
0.69
0.27
0.04
0.52
0.60
0.10
0.19
Married
Female
Not White
Disabled
Notes: Sample: All persons aged 70 and older in each year of the National
Health Interview Survey. Number of observations: 6,724 in 1982 and 4,921 in 1996.
and only 0.8% for 80- to 89-year-olds. The oldest old (90
and older) experienced improvements that were similar in
magnitude to elderly persons in their 70s (i.e., 1.3% annu-
ally). Elderly persons who were currently married experi-
enced improvements that were a full 1 percentage point
higher—1.7% versus 0.7%—than elderly persons who were
not married. Regional differences were substantial. The
West, where 17% of the sample lived in 1982, experienced
no decline in disability. The Midwest had a decline of 0.8%
on an annual basis. The Northeast (1.5% decline) and the
South (2.0% decline) had the largest improvements.
Perhaps the most striking differences were across educa-
tion groups. There were no improvements for elderly per-
sons with 0–8 years, 9–11 years, or 12 years of schooling.
The only group to experience improvements was elderly
persons with more than a high school degree. For these rela-
tively well-educated aged persons, disability declined by
1.5% on an annual basis.
Have the declines been persistent?—Figure 1 suggests
that the improvements were concentrated in the early 1980s,
with little or no subsequent improvements. Additional sup-
port for this point is provided in Table 7. First, disability
prevalence in each year relative to 1982 (in terms of odds
ratio) is reported in column 1 (“Year Dummies”). These es-
timates are the logistic regression analog to Figure 1, and
they demonstrate that most of the change occurred between
1982 and 1986. For example, the odds ratios are the same in
1987 (0.863) and 1996 (0.863). The next set of estimates
(“Year Dummies With Controls”) simply adds controls for
age, gender, and proxy reporting. Even after controlling for
these factors, the changes are concentrated in the early
1980s; the odds ratios in 1987 and 1996 are similar at 0.884
and 0.858, respectively.
Do sociodemographic factors confound the time
trend?—Table 5 reports the socioeconomic composition of
the elderly population in 1982 versus 1996. The elderly
Table 6. Trend in Disability After Adjusting for Sociodemographic Factors
Model 1—Baseline
OR 95% CI
Model 2
95% CI
Model 3
95% CI
Model 4
95% CI
Model 5
95% CI
OR
OR
OR
OR
Trend
Age
Female
Full proxy
Part proxy
DK proxy
Education: 9–11 years
Education: 12 years
Education: 13 or more
Married
0.988
1.106
1.690
2.028
1.530
1.320
0.984–0.991
1.103–1.109
1.638–1.743
1.951–2.108
1.442–1.622
1.144–1.523
0.996
1.100
1.717
1.982
1.487
1.410
0.773
0.616
0.571
0.992–0.999
1.097–1.102
1.663–1.773
1.904–2.063
1.400–1.579
1.183–1.680
0.740–0.808
0.593–0.639
0.548–0.596
0.989
1.099
1.463
2.144
1.647
1.281
0.985–0.992
1.096–1.101
1.413–1.513
2.062–2.230
1.553–1.747
1.110–1.478
0.987
1.107
1.684
1.998
1.511
1.314
0.983–0.990
1.105–1.110
1.632–1.737
1.921–2.077
1.425–1.603
1.138–1.517
0.995
1.095
1.524
2.058
1.568
1.399
0.803
0.658
0.614
0.745
1.406
0.991–0.998
1.092–1.098
1.471–1.577
1.976–2.144
1.475–1.666
1.173–1.668
0.769–0.839
0.633–0.684
0.588–0.640
0.720–0.771
1.347–1.468
55%
0.691
0.668–0.714
7%
Not White
% Change in trend from baselinea
1.640
1.574–1.709
–11%
65%
Notes: OR ꢁ odds ratio; CI ꢁ confidence interval; DK ꢁ do not know. Sample: All persons aged 70 (71 in 1982) and older in the National Health Interview Sur-
vey, 1982–1996.
aFor example, for Model 2, the percentage change is: [(0.9958ꢂ0.9880)/(1ꢂ0.988)] ꢀ 100. Number of observations ꢁ 124,949.

TRENDS IN OLD-AGE DISABILITY
S215

S216
SCHOENI ET AL.
An alternative way to depict these results is to reestimate
the baseline model but restrict the sample to the more recent
period. These estimates are reported in the last columns of
Table 7. Disability declined by 3.4% on an annual basis
over the period 1982–1986. When the sample is restricted to
the 1986 to 1992 period, the time trend is eliminated (odds
ratio of 1.002). But during the subsequent 5-year period
1992–1996—the same period over which Waidmann and
Liu (2000) analyze the MCBS—the NHIS shows an annual
decline in disability of 2.5%. Therefore, the improvements
in disability did not persist throughout the period; they were
concentrated in 1982–1986 and 1992–1996.
The time trends are also sensitive to the indicator of dis-
ability that is analyzed. The share of the elderly population
with a personal care disability has changed very little over
the entire 1982–1996 period, holding steady at roughly
8.0% (Table 1). The first set of regressions in Table 8 dem-
onstrate this point by replicating the basic model but using
personal care disability instead of any disability as the out-
come. For the 1982–1996 period, there is no trend, with an
odds ratio of 1.000. Similarly, there is no trend over the
1982–1986 period. There is actually a marginally signifi-
cant increase in personal care disability during both the
1986–1992 and 1992–1996 periods.
The observed trend in any disability (Table 4) is driven
by the improvements in routine need disability without per-
sonal care disability. The regression estimates in Table 8
show large declines of 4.4% annually over the 1982–1986
period. The gains over the 1986–1992 period are much
smaller at 0.6% annually, and they are not statistically sig-
nificant. However, the gains during the subsequent 5
years—4.9% annually—were just as large as they were over
the 1982–1986 period.
To summarize, the trends are sensitive to the measure of
disability that is analyzed, with most of the improvements at
lower levels of disability (i.e., routine need disability without
personal care disability). Moreover, improvements were not
monotonic throughout the 1982–1996 period, with most of
the decline occurring in the periods 1982–1986 and 1992–
1996, nor do they appear to have accelerated over time.
DISCUSSION
A growing body of evidence suggests that disability rates
have been declining among older Americans. This evidence
is based on analyses of a handful of surveys. These surveys
measure disability in only a few years (NLTCS, SIPP,
SOA), or they are available for a relatively short time span
(MCBS). The only exception is the NHIS, which allows an-
nual estimates over a 15-year period, 1982–1996. This
study analyzed the most recent data from the NHIS, and it
integrated these analyses with evidence from other national
surveys. Four questions were addressed:
• Are the declines persistent over a long period of time?
• Are the findings consistent across surveys?
• Are the trends widespread, or are they limited to certain
sociodemographic groups?
• Are the trends robust to various specifications?
The weight of the evidence implies that declines did not
persist throughout the entire 1982–1996 period. There were

TRENDS IN OLD-AGE DISABILITY
S217
clear decreases in the prevalence of disability between 1982
and 1986, but no improvements between 1986 and (at least)
1992. Disability began to decline again more modestly
around 1992, falling through 1996 (the last year of available
data). The estimates from the NHIS also demonstrate that
any improvements that occurred throughout the period were
for those who only need help with routine care activities—
such as everyday household chores, doing necessary busi-
ness, shopping, and getting around—rather than the more
severe indicator of personal care disability, which demon-
strated no change over the 1982–1996 period. The evidence
is fairly consistent across five national surveys.
The trends observed in this study do not appear to be sen-
sitive to the exclusion of the nursing home population from
the NHIS; however, it is less clear to what extent trends in
assisted living and other residential care arrangements may
be contributing to improvements in the noninstitutionalized
population. This important component of long-term care is
often not sampled in most community-based household sur-
veys and also excluded from the national nursing home sur-
vey efforts. Other surveys (such as the NLTCS and MCBS)
include Medicare beneficiaries wherever they reside, in-
cluding assisted living communities, but do not contain
large enough samples of assisted living residents to pursue
this important question. Given the rapid expansion of this
industry over the past decade, and the high likelihood of
continued growth to meet the needs of the aging population,
survey designers should consider including and oversam-
pling assisted living residents in future survey efforts.
We also find that improvements were experienced rela-
tively widely across the population: men and women,
Whites and non-Whites, most age groups, and married and
unmarried. However, within education groups, improve-
ments were concentrated among the most educated. There
were no improvements in disability among people with 12
or fewer years of schooling. Although the educational at-
tainment of the older population has increased over time
and will continue to increase over the coming decades (U.S.
Bureau of the Census, 1996), as a society we need to under-
stand why those who have not completed high school are
not benefiting from what otherwise appears to be a general
experience.
cise, smoking patterns, access to technology and assistive
devices, and access to more facilitative environments when
disability occurs. Given the consistency of results across
surveys with respect to the importance of education, a better
understanding of the pathways through which education af-
fects late-life disability may be a useful place to start in ex-
panding our understanding of disability trends.
To more fully understand the changes in disability and
identify the factors causing the improvements, better data
are needed on the underlying processes related to disability
prevalence: incidence, recovery, and survival. Expansion of
the abbreviated ADL and IADL disability questions in the
1992–1996 NHIS to include a more detailed battery of
items (as has been done beginning in 1997) is a useful first
step, as is linking the NHIS to mortality data. Annual fol-
low-up interviews with older NHIS respondents, along the
lines of the LSOA, would also be a useful design augmenta-
tion. Supplementation of self- and proxy-reported items
with physical performance measures, which may evaluate
capacity in a more standardized manner over time, could
help minimize issues related to item consistency and inter-
pretation in future studies of disability trends.
Until we understand the causes of disability declines
among older Americans, the implications of those improve-
ments cannot be fully assessed. If improved access to and
use of clinical care is driving the trend, the budgetary impli-
cations for the future of Medicare could be enormous. Alter-
natively, if environmental and technological changes—such
as more widespread use of microwaves to facilitate meal
preparation, ramps and curb cuts to facilitate going outside,
and computers to facilitate shopping on-line—underlie im-
provements in routine care functioning, then the outlook for
Medicare is likely to be somewhat less ominous. A fuller
understanding of the causes is also important for planning
and implementing future public health efforts. If the im-
provements are related to environmental changes, for exam-
ple, then practitioners and program designers may be led to
focus on very different interventions than if the improve-
ments are related to changes in preventive and hygienic fac-
tors or in the use of medical care. Clearly, determining the
causes of the improvements should be a high priority in fu-
ture research efforts.
More important, given the evidence that disability is still
declining in the 1990s, we clearly need a better understand-
ing of why improvements are occurring. Such an under-
standing is critical for planning for the long-term care and
social service needs of the older population. Although we
did not attempt causal explanations in this study, our find-
ings that (a) only the most educated have experienced de-
clines; (b) the declines occurred for routine care rather than
personal care needs; and (c) education is the most salient so-
ciodemographic confounder of the time trends offer some
insights into possible mechanisms for further exploration.
As others have noted (Freedman & Martin, 1999; Waid-
mann & Liu, 2000), education in this context is not simply
reflective of years in the classroom; instead, it is a broad
marker for socioeconomic status, which may operate on dis-
ability through a number of pathways including access to
medical care and patterns of medical care utilization, as well
as health-related practices and behaviors such as diet, exer-
Acknowledgments
This work was supported by National Institute on Aging Grants
K01AG00670 to Dr. Schoeni and R29-AG14346 to Dr. Freedman. Michael
Hurd, Ellen Kramarow, and Jim Smith provided valuable comments.
Address correspondence to Dr. Robert F. Schoeni, RAND, 1700 Main
Street, Santa Monica, CA 90407. E-mail: schoeni@rand.org
References
Agree, E. M., & Freedman, V. A. (1999). The implications of population
aging for geriatric health. In J. J. Gallo (Ed.), Clinical aspects of aging
(5th ed., pp. 426–450). Baltimore: Williams & Wilkens.
Bishop, C. E. (1999). Where are the missing elders? The decline in nursing
home use, 1985 and 1995. Health Affairs, 18, 146–155.
Clark, D. O. (1997). US trends in disability and institutionalization among
older blacks and whites. American Journal of Public Health, 87, 438–
440.
Colvez, A., & Blanchet, M. (1981). Disability trends in the United States
population: 1966–1976: Analysis of reported causes. American Journal
of Public Health, 71, 464–471.

S218
SCHOENI ET AL.
Crimmins, E. M., Saito, Y., & Reynolds, S. L. (1997). Further evidence on
recent trends in the prevalence and incidence of disability among older
Americans from two sources: The LSOA and the NHIS. Journal of
Gerontology: Social Sciences, 52B, S59–S71.
Dey, A. N. (1997). Characteristics of elderly nursing home residents: Data
from the 1995 National Nursing Home Survey. Advance Data from Vi-
tal and Health Statistics (No. 289). Hyattsville, MD: National Center
for Health Statistics.
Freedman, V. A., & Martin, L. G. (1998). Understanding trends in func-
tional limitations among older Americans. American Journal of Public
Health, 88, 1457–1462.
Freedman, V. A., & Martin, L. G. (1999). The role of education in explain-
ing and forecasting trends in functional limitations among older
Americans. Demography, 36, 461–473.
Freedman, V. A., & Martin, L. G. (2000). The contribution of chronic con-
ditions to aggregate changes in old-age functioning. American Journal
of Public Health, 90, 1755–1760.
Freedman, V. A., & Soldo, B. J. (Eds.). (1994). Forecasting disability:
Workshop summary. Convened by the Committee on National Statis-
tics of the Commission on Behavioral and Social Sciences and
Education, National Research Council. October 4–5, 1993. Washing-
ton, DC: National Academy Press.
Hawes, C., Rose, M., & Phillips, C. D. (1999). A national study of assisted
living for the frail elderly: Results of a national survey of facilities.
Beachwood, OH: Myers Research Institute.
Hing, E., Sekscenski, E., & Strahan, G. (1989). The National Nursing
Home Survey: 1985 summary for the United States. Vital and Health
Statistics Series 13, no. 97 (DHHS Publication No. PHS 89-1758).
Washington, DC: U.S. Government Printing Office.
in elderly United States populations: 1982–1994. Proceedings of the
National Academy of Sciences, 94, 2593–2598.
National Center for Health Statistics. (1999). Health, United States, 1999
with health and aging chartbook. Hyattsville, MD: Author.
Rhoades, J. A., & Krause, N. A. (1999). Nursing home trends, 1987 and
1996 (MEPS Chartbook No. 3). Rockville, MD: Agency for Health
Care Policy and Research.
Spector, W. D., Fleishman, J. A., Pezzin, L. E., & Spillman, B. C. (1998).
The characteristics of long-term care users. Washington, DC: Institute of
Medicine, for the Committee on Improving Quality in Long-Term Care.
U.S. Bureau of the Census. (1996). 65ꢃ in the United States. Current Pop-
ulation Reports, Special Studies, P23-190. Washington, DC: U.S.
Government Printing Office.
U.S. Bureau of the Census. (1997). Listing and coverage manual for field
representatives. Washington, DC: U.S. Government Printing Office.
U.S. Bureau of the Census. (2000). Population estimates of the United
States resident population [On-line]. Available: http://www.census.
gov/population/www/estimates/uspop.html
Verbrugge, L. (1984). Longer life but worsening health? Trends in health
and mortality of middle-aged and older persons. Milbank Memorial
Fund Quarterly, 62, 475–519.
Waidmann, T., Bound, J., & Schoenbaum, M. (1995). The illusion of fail-
ure: Trends in the self-reported health of the U.S. elderly. Milbank
Memorial Fund Quarterly, 73, 253–287.
Waidmann, T., & Liu, K. (2000). Disability trends among the elderly and
implications for the future. Journal of Gerontology: Social Sciences,
55B, S298–S307.
Wilson, R. W., & Drury, T. F. (1984). Interpreting trends in illness and dis-
ability: Health statistics and health status. Annual Review of Public
Health, 5, 83–106.
Manton, K. G., Corder, L., & Stallard, E. (1993). Estimates of change in
chronic disability and institutional incidence and prevalence rates in the
U.S. elderly population from the 1982, 1984, and 1989 National Long
Term Care Survey. Journal of Gerontology: Social Sciences, 48, S153–
S166.
Received July 31, 2000
Manton, K. G., Corder, L., & Stallard, E. (1997). Chronic disability trends
Accepted December 19, 2000