Overview of the Health Measures in
the Health and Retirement Study
Robert B. Wallace
A. Regula Herzog
ABSTRACT
This report discusses the conceptual and logistical issues that lead to
the items and instruments used to measure health and function status in
Wave 1 of the Health and Retiremeni Study (HRS). Health status was
conceptualized as multidimensional, and included selected major symptoms, diseases and conditions, global self-assessment of health, physical
and cognitive functional status, the utilization of health services and selected elements of emotional health. In addition, two physiologic measures were obtained on a sample basis: grip strength and pulmonary
maximum expiratory flow rate. Prevalence rates for major conditions
and functional states are presented. Correlational analyses generally
demonstrated a high rate of convergent, discriminant and construct
validity. These findings should assist those intending to analyze HRS
data in terms of the focus and utility of the measures employed.
I. Introduction
A major goal of the Health and Retirement Study (HRS) is to
explore the role of health in the retirement decision and the long term health
consequences of the retirement process. More generally, the health measures
make possible investigations ofthe interface between health and economic behaviors and status in the later part of life. A comprehensive conceptualization of the
relationship between retirement and health views both as processes that condition
each other over a large part of the life cycle. Antecedent health problems and
their environmental causes may have their roots relatively early in the individual
Robert B. Wallace is a professor of preventive medicine at the University of Iowa, A. Regula Herzog
is a research scientist at the University of Michigan. This research was supported by a Cooperative
Agreement with the National Institute on Aging AG09740. The data used in this article are from the
alpha release of the HRS.
THE JOURNAL
OF HUMAN RESOURCES
' XXX • Supplement
1995
Wallace and Herzog
life course. Depending on the timing and intensity of developing physiologic and
functional impairments during the life course, these may help determine the
amount and nature of employment. While the HRS cannot explore the remote
origins or early history of most health, employment, or retirement activities, it
strives to characterize these features in the baseline age range of 51 to 61 years
of age and to follow their trajectories.
Health assessment within the HRS was designed to capture respondent health
and functional status and medical care utilization as fully as possible, realizing
the constraints of survey time and resources available, and to be maximally responsive to the major study hypotheses and goals. The health measurements for
Wave 1 were developed by the Health Working Group.' In developing the health
measures the group consulted many prior health surveys and scientists with relevant expertise. This paper considers selected health measurement issues relevant
to the study of health and retirement, reviews the major strategies that guided
the selection of the specific health measures, presents basic health measures
utilized in the HRS baseline survey and resulting distributions, and explores
statistical relationships among important health measures in order to evaluate
validity and guide future analytic efforts.
II. Health Measurement Strategies in the HRS
In the HRS, health was conceptualized as multidimensional, with
a general emphasis on physical and mental (including cognitive) domains. Items
on social health and function can be found throughout other sections of the survey
instrument, covering such areas as the work and family environment. Physical
health was viewed as multiaxial (Cote 1982), attempting to capture as many axes
as possible, including both subjective and objective assessments. This will allow
the analyst to apply various conceptual frameworks of personal interest. General
health measure categories included: 1) self-rated overall health status and recent
changes in overall health, 2) the presence and severity of common chronic medical
conditions, 3) important symptoms and syndromes, 4) function status measures,
5) previously established important hygienic behaviors, 6) utilization of general
health services and selected medical treatments and social support services, and,
in an experimental subsample, 7) two physiological performance measures: handgrip strength and maximum peak expiratory flow rate. It was not the intent of
HRS to generally assess the causes of various diseases and dysfunctions present
at baseline.
These dimensions are consistent with other comprehensive health assessment
instruments. For example, the comprehensive and well-evaluated Sickness Impact Profile (Bergner et al. 1981; Williams 1994) contains questions on physical,
cognitive, emotional, and social functioning. The thoroughly evaluated SF-36,
derived from the Medical Outcomes Study (McHorney, Ware, and Raczek 1993;
I. The Health Working group included John Bound. Regula Herzog (coordinator). Mary Grace Kovar.
Jersey Liang. Willard Manning. Willard Rodgers, Frank Sloan. Robert Wallace (chair), and John Ware.
S85
S86
The Journal of Human Resources
Stewart and Ware 1992; Wright 1994), includes most of these categories as well
as items on symptoms and overall self-ratings of health. Although the choice and
format of specific questions vary from one measure to the next, most available
health assessment tools, including the HRS, derive from a similar conceptual
and measurement tradition. According to that tradition, health is viewed as a
multidimensional concept that includes social, behavioral, and psychological dimensions and health perceptions, in addition to medical pathophysiology and
clinical phenomenology. The functional dimension of health has been articulated
in the Institute of Medicine's (Pope and Tarlov 1991) disability model of health
which, in turn, is grounded in the World Health Organization's International
Classifications of Impairments, Disabilities, and Handicaps and in the functional
limitation framework proposed by Nagi (1965). The subjective or evaluative dimension of health perceptions derives from the increasing emphasis on quality
of life and clinical outcomes in medical research (Patrick and Erickson 1993) and
the emphasis in those measures on subjective perception and expectations of
health.
A general issue faced when formulating HRS health measures was whether to
employ individual items or item sets from diverse sources or to use multidimensional instruments already in existence. Each approach has strengths and weaknesses. The use of existing instruments, such as the SF-36, the Sickness Impact
Profile, the Cornell Medical Index or many available disease-specific health instruments would give the benefit of prior experience and information on measurement properties, and would allow comparison with many studies and crossreferences with various national or local studies. Most also presented problems
for the HRS, however, including excessive length, undue attention to the more
severe end of the health and functional spectrum, failure to address various health
domains such as diseases or health care utilization, lack of discrimination of small
variations in health or functional status postulated to relate to work performance,
and the lack of inclusion of recently described, promising items.
The process of designing and designating items and scales for inclusion was
complex and iterative. There was extensive literature review and scientific consultation to be certain that the menu for item selection was broad and modern.
Where raw data on selected items were available from previous surveys, analyses
were conducted to assess reliability and construct validity, reduce redundancy
and evaluate desired associations with other variables. The three survey pretests
resulted in alterations of initial selection and wording. Several criteria were used
in designing or otherwise specifying health items, although it was clearly not
possible to meet all criteria at all times:
• High quality measures that can provide maximally valid and reliable information for the HRS.
• Proven items and instruments from the scientific literature, counterbalanced by the need to create new items responsive to modern hypotheses
and conceptualizations of health and its interrelationship with work.
• Items and instruments that are at least in part represented in other recent
national surveys in order to compare and contrast findings.
Wallace and Herzog
• Measures that can be administered repeatedly and be sensitive to subtle
changes in health status.
• Measures that are appropriate for persons of various cultural and socioeconomic backgrounds, particularly persons of Hispanic and AfricanAmerican backgrounds.
• Measures that are adaptable for the telephone administration of future HRS
waves, necessitating items that are simple and well understood, with short
response scales that can be used without visual aids.
• Level of item difficulty that is suitable for participants with both high and
low functional status.
In addition, past research on retirement and health suggested several considerations for framing the HRS health measures.
A. Linked Versus Independent Health Assessment
Whereas existing econometric studies have generally confirmed the significance
of health in retirement behavior, the magnitude of effect varies (for reviews see
Chirikos 1993; Quinn and Burkhauser 1990). One reason may relate to the health
measures utilized. For example, self-attribution of work limitation or retirement
to health factors may suggest a stronger role for health status than independently
measured disease and disability factors which are then analytically linked to retirement behavior. Which type of measure yields the more accurate effect has
been a matter of considerable debate. The former approach may produce overstated effects because they allow for easy rationalizations (Anderson and Burkhauser 1984, 1985; Bound 1991; Myers 1982), but the latter approach may produce
understated effects because it assesses aspects of health that are not directly
relevant to the work setting at hand. The HRS included both types of measures.
B. Global Versus Specific Health Measures
Most prior research on health and retirement emphasized global measures of
health; research that examined the effect of specific conditions or symptoms
(Bartel and Taubman 1979; Burkhauser, Butler, and Mitchell 1986; Mitchell and
Anderson 1989) typically did not include comprehensive health measurement. In
an attempt to address this deficiency, the HRS health measurement includes a
good range of specific illnesses, functions, and symptoms.
C. Objective Versus Self-Reported and Subjective Health Measures
An issue permeating research on the effect of health in the retirement decision
relates to the self-reported or even subjective nature of many of the health measures utilized in this research, and the assumption that more objective health
measures would provide better estimates because they are not as affected by the
possibly biased or unreliable reporting of individuals. Opponents of this assumption point to the respectable correlation between self-reported health status and
S87
S88
The Journal of Human Resources
assessments provided by health care professionals (for example, Ferraro 1980; La
Rue et al. 1979; Maddox and Douglas 1973). In our view, subjective perceptions of
health have independent value because they capture knowledge and interpretations not reflected in more objective indicators. The retirement decision is in part
an individual decision, and subjective perceptions—regardless of whether they
are accurate reflections of objective indicators—are likely to play a role in this
decision. Thus, the HRS contains several subjective measures as well as more
objective health measures such as specific disease conditions, symptoms, and
function.
A companion issue was whether to include clinical or physiologic tests to complement the self-report information. For several reasons the Health Working
Group eventually decided against full sample physiologic or physical performance
testing in the baseline survey, including insufficient survey time and resources,
the logistical difficulties of interviewing in the home and the knowledge that
subsequent waves would be by telephone. As a compromise, two physical performance tests were included in an experimental module, described below. Data
from the module could then be used to calibrate self-report against performance
information.
D. Cognitive and Emotional Health
Much of the previous research on health and retirement conceptualized health in
the physical, organic sense or as general health, but rarely as cognitive health.
Yet, today's occupations often require substantial cognitive competence. Currently, 58 percent of all jobs are in professional, managerial, technical, sales, and
administrative areas (U.S. Bureau ofthe Census 1994). Also, somewhat neglected
in prior research were mental disorders, common causes of work disability. For
example, mental or nervous disorders were cited as primary diagnoses for more
than 10 percent of 50-61 year old Social Security beneficiaries who were first
entitled to Disability Insurance (Hennessey and Dykacz 1989); this is almost
certainly an underestimate because of unreported and undiagnosed mental problems. One ofthe innovative features of HRS is a broadening ofthe conceptualization of health to provide for separate assessments of emotional health and cognitive functioning in addition to physical health.
II. Methods Used in this Report
The following is an overview of the health measures obtained in
the HRS, with selected review ofthe item categories and items chosen and discussion of certain item limitations. Two different types of analyses are pursued and
briefly reported in this paper, intended to probe the quality and relevance of the
health measures for studying the retirement decision and process. The first type
offers univariate distributions and missing data levels. The second type probes
various forms of reliability and validity. These analyses establish, in part, internal
consistency for some multi-item indices. We examine concurrent and discriminant
validity between various concepts. We investigate construct validity by exploring
Wallace and Herzog
theoretically predicted or empirically well-established relationships using the
HRS health variables. Data presented here are from the Gamma Release data
tape. The analyses are limited to persons 51-61 years of age, including spouses
of primary respondents within this age range. Significance calculations do not
take account of the complex sampling design.
This report contains a series of tables to introduce the analyst to selected
health item prevalences and distributions and their associations with other health
variables. These variables and the tables are discussed further below. Table 1
shows the reported frequencies for major conditions studied. Tables 2, 3, and 4
examine respectively the bivariate relations between these conditions and function status measures, cognitive and emotional measures and health care utilization. Table 5 shows the correlations among the major health and performance
measures used in the HRS. Table 6 shows the distribution of reported difficulty
for physical function status items. Finally, Table 7 shows the correlations among
the cognitive and emotional health measures.
III. Physical Health Measures
A. Specific Clinical Conditions
In the HRS, important chronic diseases were emphasized and given a high priority, as they have characteristic patterns of preventability and health service needs
and predictable functional outcomes and prognoses. They also often form the
administrative basis for disability designation. With constraints on interview
length, we focused on a subset of conditions which have public health significance—diseases that are most prevalent among middle aged and elderly persons
and/or which are most likely to result in work disability. By tracking these conditions in future waves, HRS investigators will be able to monitor these conditions
for their impact on disability, job mobility and job loss, retirement behavior, and
mortality.
The conditions selected were lifetime histories of hypertension, diabetes, cancer, chronic lung disease, heart problems (including angina and congestive heart
failure), stroke, psychiatric problems, and arthritis. If a respondent reported any
of the target conditions, additional disease-specific information was collected to
determine whether the respondent was currently being treated for that condition
and if so, the type of treatment and/or intensity of care that was used, and
whether it was perceived to impair various physical functions. This was done to
derive additional indicators of condition recency, activity severity, and treatment
intensity. Based on these questions all conditions but hypertension and arthritis
were differentiated into a severe and a less severe form. Some additional diseases
and conditions were queried in general categories by organ or organ system, so
that if important analytic associations emerged, the analyst could capture more
clinical and functional detail by other measures. The latter conditions include
asthma, back problems, feet and leg problems, kidney and bladder problems,
stomach or intestinal ulcers, high cholesterol, and fractures.
We queried the date of onset for a few conditions if we had evidence that the
S89
S90
The Journal of Human Resources
Table 1
Prevalence of Medical Conditions: Ever and Severe.
HRS Respondents 51-61 years old
High blood pressure
Diabetes (current -I- insulin)
Cancer (treatment/12 months)
Chronic lung disease (and activity limitations)
Heart condition (congestive heart disease and
medication or shortness of breath)
Stroke (and health consequences)
Psychiatric problem (current and medication
or treatment)
Arthritis
Any conditions
Ever had
condition
(percentage)
Severe form
(percentage)
38
10
6
8
13
not ascertained
2
2
3
I
3
I I .
38
69
I
4
not ascertained
IT'
Ni)te: Definition of severity is indicated within parentheses. The number of respondents is about
9.3(K).
a. Does not include high blood pressure and arthritis because no severe lorm of those conditions Vk'ere
a.sses.sed.
respondent could accurately recall it or if we thought knowledge of this information would aid in understanding disease severity or prognosis. For most conditions, however, we decided to monitor the clinical trajectory using information
in subsequent survey waves. The potential problems with over- and underreporting of medical conditions was addressed by careful wording of the questions. Even the clearest wording would not suffice, however, if the respondent
was unaware of a certain medical condition or did not wish to report it, and some
misclassification of respondents is therefore possible. Eventually, the HRS survey
data will be linked to Medicare claims and possibly other medical care data bases,
which will furnish independent estimates on the incidence and prevalence of
medical conditions.
Table I presents prevalences for several conditions surveyed; prevalences for
conditions "ever had" are shown in Column I, prevalences for the "severe"
conditions in Column 2. The abbreviated criteria used for defining severity depend
on the specific condition and are indicated in parentheses in the table. The most
prevalent conditions measured in HRS were arthritis and hypertension; about 40
percent of all HRS respondents reported that they had these conditions at some
point in their life. Heart problems, psychiatric problems and diabetes were somewhat less prevalent. Rates for the severe forms of the conditions ranged from
1-4 percent. Tables 2-4 relate the presence and severity of study illnesses to
other health measures. In most instances, the presence and in turn severity of
Wallace and Herzog
Table 2
HRS Physical Health Measures hy Disease Severity,
HRS RespondetUs 51-61 years old
Mobility
Difficulty
Index
(0-5)
Large
Muscle
Difficulty
Index
(0-4)
Difficulty
Index
(0-3)
Poor
Vision
(1-5)
0.79
1.31**
1.13
1.56**
0.05
0.11**
2.14
2.36**
0.91
1.53**
2.24**
1.23
1.83**
2.09**
0.06
0.13**
0.28**
2.18
2.54**
2.91**
0.97
1.14*
1.59**
1.27
1.57**
1.79**
0.07
0.08
0.17**
2.22
2.21
2.30
0.89
1.39**
3.30**
1.23
1.60**
2.70**
0.06
0.07
0.48**
2.18
2.40**
3.14**
0.87
1.66**
2.96**
1.20
1.85**
2.56**
0.05
0.16**
0.48**
2.17
2.51**
2.95**
0.95
1.57**
1.27
1.71**
0.06
0.11
2.20
2.54**
2.99**
2.65**
0.79**
3.12**
0.88
1.68**
2.25**
1.20
1.93**
2.33**
0.05
0.17**
0.35**
2.16
2.55**
2.88**
0.68
1.50**
0.93
1.91**
0.03
0.14**
2.12
2.38**
ADL
High blood pressure
No
Yes
Diabetes
No
Yes, mild
Yes, severe
Cancer
No
Yes, not active
Yes, active
Chronic lung disease
No
Yes, mild
Yes, severe
Heart condition
No
Yes, mild
Yes, severe
Stroke
No
Yes, no health consequences
Yes, health consequences
Emotional, nervous, or psychiatric problem
No
Yes, mild
Yes, severe
Arthritis
No
Yes
S9I
S92
The Journal of Human Resources
Table 2 (continued)
High blood pressure
No
Yes
Diabetes
No
Yes, mild
Yes, severe
Cancer
No
Yes, not active
Yes, active
Chronic lung disease
No
Yes, mild
Yes, severe
Heart condition
No
Yes, mild
Yes, severe
Stroke
No
Yes, no health consequences
Yes, health consequences
Emotional, nervous or psychiatric problem
No
Yes, mild
Yes, severe
Arthritis
No
Yes
High Peak
Expiratory
Flow
(0-800)
High
Grip
Strength
(0-200)
Poor
Hearing
(1-5)
Usual
Pain
(0-3)
Worst
Pain
(0-3)
2.20
2.39**
0.33
0.51**
0.49
0.73**
429.35
422.83
83.08
81.35
2.24
2.50**
2.61**
0.37
0.60**
0.85**
0.54
0.84**
1.15**
427.59
419.80
83.52
77.26
2.28
2.19
2.25
0.39
0.55**
0.67**
0.57
0.75**
0.95**
429.62
414.24
+
83.29
70.45*
+
2.24
2.48**
2.88**
0.36
0.58**
1.23**
0.53
0.88**
1.66**
434.26
355.58**
307.48**
82.58
88.77
69.80
2.23
2.53**
2.86**
0.35
0.52
0.68** 0.94**
1.22** 1.63**
428.43
421.37
83.23
80.13
2.26
2.64**
0.38
0.44
0.57
0.75
427.09
2.76**
1 19**
1.50**
2.23
2.51**
2.74**
0.32
0.49
0.85** 1.24**
1.26** 1.54**
429.73
394.17
417.96
83.32
71.61*
82.16
2.16
2.45**
0.18
075**
438.78
409.63*
86.69
75.74**
0.28
1.07**
+
+
+
+
82.70
+
+
+
+
Note: Column variables are coded in the direction of the label. Entries are means: statistical significance (*p < .05, **p < .01) refers to the difference between the specific severity level and the level
without the disease. For definition of severity see Table I. The number of respondents varies between
9,400 and 9,700 except for grip strength and expiratory fiow rate which were answered by 530 respondents in an experimental module. Cells labeled with + contain less than 15 respondents.
Wallace and Herzog
Table 3
HRS Mental Health Measures by Disease Severity.
HRS Respondents 51-61 years old
High blood pressure
No
Yes
Diabetes
No
Yes. mild
Yes, severe
Cancer
No
Yes, not active
Yes, active
Chronic lung disease
No
Yes, mild
Yes, severe
Heart condition
No
Yes, mild
Yes, severe
Stroke
No
Yes, no health consequences
Yes, health consequences
Emotional, nervous, or psychiatric problem
No
Yes, mild
Yes, severe
Arthritis
No
Yes
Immediate
Word
Recall
(0-20)
Delayed
Word
Recall
(0-20)
WAIS-R
Similarities
(0-14)
IADL
Difficulty
Index
(0-3)
Low
CESD-ll
(1-4)
7.75
7.39**
5.71
5.27**
6.56
6.05**
0.44
0.54**
3.60
3.48**
7.68
7.18**
6.67**
5.61
5.12**
4.51**
6.44
5.85**
5.20**
0.46
0.57**
0.81**
3.57
3.47**
3.30**
7.61
7.94*
7.46
5.55
5.63
5.14
6.37
6.43
6.16
0.47
0.48
0.52
3.56
3.53
3.46**
7.66
7.37*
6.80**
5.58
5.37
4.71**
6.41
6.21
5.20**
0.46
0.48
0.85**
3.58
3.49**
3.03**
7.69
7.22**
6.54**
5.61
5.18**
4.56**
6.42
6.09**
5.23**
0.47
0.51
0.72**
3.58
3.43**
3.16**
7.64
7.49
5.56
5.42
6.39
5.95
0.47
0.47
3.57
3.43**
6.15**
4.33**
5.21**
1.10**
3.15**
7.68
7.31**
6.68**
5.60
5.31*
4.64**
6.41
6.17
5.78**
0.44
0.70**
0.89**
3.61
3.28**
2.91**
7.66
7.54*
5.58
5.48
6.47
6.20**
0.44
0.54**
3.62
3.46**
Note: Column variables are coded in the direction of the label. Entries are means; statistical significance (*p < .05, **p < .01) refers to the difference between the specific severity level and the level
without the disease. For definition of severity see Table 1. The number of respondents varies between
8,600 and 9,700.
S93
S94
The Journal of Human Resources
Table 4
HRS Health Care Measures by Disease Severity.
HRS Respondents 51-61 years old
Number
of Times
Hospitalized
High blood pressure.
No
Yes
Diabetes
No
Yes, mild
Yes, severe
Cancer
No
Yes, not active
Yes, active
Chronic lung disease
No
Yes, mild
Yes, severe
Heart condition
No
Yes, mild
Yes, severe
Stroke
No
Yes, no health consequences
Yes, health consequences
Emotional, nervous or psychiatric problem
No
Yes, mild
Yes, severe
Arthritis
No
Yes
Number
Number
Self Care
of Doctor
of Days
Visits
Home Sick Minutes
0.11
0.21**
2.29
3.52**
0.95
1.38**
3.76
6.72**
0.13
0.26**
0.56**
2.60
3.89**
5.22**
1.05
1.52**
2.21**
4.32
7.92**
17.47**
0.14
0.17
0.78**
2.67
3.44**
5.52**
1.08
1.43**
2.36**
4.63
6.54
14.65**
0.13
0.19**
0.66**
2.64
3.38**
4.96**
1.02
1.48**
3.35**
4.14
5.88
25.38**
0.11
0.36**
0.91**
2.55
3.96**
5.33**
1.00
1.76**
2.99**
3.77
9.91**
32.56**
0.14
0.35**
2.71
3.91**
1.08
1.54*
4.53
10.52*
0.77**
5.07**
3.25**
24.43**
0.13
0.21**
0.46**
2.58
3.49**
5.37**
0.99
1.77**
2.82**
3.69
9.34**
24.14**
0.12
0.20**
2.33
3.46**
0.87
1.52**
3.24
7.60**
Note: Column variables are coded in the direction of the label. Entries are means; statistical significance (*p < .05, **p < .01) refers to the difference between the specific severity level and the level
without the disease. For definition of severity see Table 1. The number of respondents is about 9,700.
Wallace and Herzog
the conditions was associated with worsening health, for example in the form of
pain, mobility, and physical dysfunction.
B. Symptoms
A deliberate decision was made not to comprehensively measure symptoms because they are often associated with specific diseases, and because of space
constraints in the survey instruments. Two important symptoms—pain and decreased stamina or vitality—were surveyed, however, because they are common
and often occur without specific disease or mechanistic attribution. Both can
interfere with performance on the job and other physical and social functions.
Three measures of the prevalence of frequent pain and its intensity were chosen.
In the HRS, 24 percent of the respondents reported being troubled by pain; and
16 percent of these rated their average pain as severe, 53 percent their worst pain
as severe. The medical conditions associated with most severe pain are severe
lung disease and heart condition (Table 2). The two pain measures are also
strongly associated with functional difficulty in terms of mobility and large muscle
strength (Table 5). Three items assessing personal vitality were taken from the
baseline Established Populations for Epidemiologic Study of the Elderly questionnaire (Huntley et al. 1986). All these have been found to be predictive of chronic
illness in both cross-sectional and longitudinal analyses (Mobily et al. 1994). The
HRS also contains other selected symptom reports in various scales, such as
restless sleep in the depression symptom inventory.
C. Functional Status
Another important dimension of health is functional status, the ability to perform
various defined tasks, usually in an implicitly or explicitly specified environmental
context. Higher functional levels are likely to be critical elements in maintaining
employment. Because of resource and logistical limitations, the function status
measures chosen for HRS baseline are all based on self-report. The strengths
and weaknesses of self-reported versus performance-based measures have been
reviewed (Guralnik and LaCroix 1992). In general, function status measures summarize overall health status, including age-related anatomic, physiologic and cognitive changes as well as prevalent clinical conditions. They were selected to
meet several criteria: to cover a broad range of basic and complex tasks that
might be found in a general community-dwelling population; to represent upper
and lower body mobility and fine motor skills; to differentiate across clinical
conditions of a diverse nature and severity; to reflect varying levels of physical
endurance; and to be responsive to change on repeated measurement. The tasks
queried in the survey instrument included basic activities of daily living (ADLs),
as well as more physically or cognitively demanding tasks such as running a mile
or using a computer. Individual movements, such as lifting, stooping and kneeling, or picking up a small object were incorporated to assess their relation to
performance on specific jobs and the retirement decision. We used a four-point
response scale (not at all difficult, a little difficult, somewhat difficult, very difficult/can't do). The prevalences for difficulty with various function status mea-
S95
S96
The Journal of Human Resources
I
I
I
I
I
I
I
I
I
I
I
00
-*
OS 0 0 - ^
• * <N —
I
I
I
I
I
I
o
ON «O <NI
— —• OS
2
'S.
(^ ^3 lO
— fN ^-
I
o
00 r^ r^ •—' (N
— ^- — ,—. rn
*
*
*
*
oo
(N
* *
* *
o<N CN
m
*
*
*
•X*
( N *O 0 0
<N
I
*
8
*
*
m
s
00
o
2
o.
X
•73
n
3
X
X
•a
2 il
c
"3 (5 t!,£" I
S "o "3
i.i
>• 3> 5
_ _^ 5
Q. ^ ^
X XI
.uS.Se n•— . ^ oil
C
ca
(U
ca
Q.
-C _
cd
D.
«
ed
<U
13
u
ca
I
0) i±2
1- • ^
£^ w
<"
;5 faOi-J
3 fc
o SQ
S J <
(£ D ^ X a c2
oo OS — —
Is
. c
V 'Z,
Wallace and Herzog
Table 6
Prevalence of Functional Difficulties. HRS Respondents 51-61 years old
(in percentages)
Run or jog a mile
Walk several blocks
Walk one block
Walk across a room
Sit for two hours
Get up from a chair after sitting long
Get in and out of bed without
help
Climb several flights of stairs
Climb one flight of stairs
Lift or carry weights over 10
pounds
Stoop, kneel or crouch
Pick up a dime from the table
Bathe or shower without help
Lift arms above shoulder
Pull or push large objects
Eat without help
Dress without help
Use a map in a strange place
Use a microwave after reading instruction
Use a calculator to balance
checkbook
Use a computer or wordprocessor
Not at All
Difficult
A Little
Difficult
Somewhat
Difficult
Very
Difficult/
Can't Do
17
77
90
96
71
65
21
10
5
2
14
21
23
7
3
1
9
10
39
6
2
1
6
4
93
4
2
1
53
22
8
11
12
6
13
4
6
96
89
75
99
97
70
90
21
3
2
6
12
1
2
16
7
11
2
1
3
6
0
1
7
2
10
1
1
2
7
0
0
7
1
90
6
2
2
57
16
11
16
83
77
58
94
5
Note: The number of respondents varies between 9.700 and 5,850.
sures are shown in Table 6. A few questions (for example, jogging) yield a nonnegligible number of missing data. An exploratory factor analysis of the HRS physical
functioning items indicates three distinct factors: (a) a mobility factor including
all ambulation except the jogging items and all stair-climbing items; (b) a large
muscle strength factor including lower as well as upper extremities (namely, sit
for about two hours; get up from a chair after sitting for long periods; stoop,
kneel, or crouch; pull or push large objects like a living room chair); and (c) an
ADL-related factor. Three indices were formed as simple counts of the number
S97
S98
The Journal of Human Resources
of items in each factor with which the respondent reported any difficulty. If these
function status measures are valid, they must relate to other health measures in
theoretically meaningful ways. This is just what we found in initial analyses of
the HRS. For example, as shown in Table 2, the presence of all eight diseases is
associated with more functional problems and in most instances the difference is
statistically significant.
D. Sensory Function
Because sense organ deficits are among the most prevalent functional impairments in middle and older ages and are among those with the earliest onset, items
on vision and hearing functions were included. These functions were queried
under "best-corrected" conditions, that is, with whatever glasses, contact lenses,
hearing aids, or other devices the respondent was using. As shown in Table 5,
vision relates to functioning in terms of mobility as well as to other health .measures, while hearing relates generally less strongly to all health measures.
E. Physiologic Measures
Two physiologic tests were performed in an experimental module, in order to
explore global health status and vitality using simple, portable devices. Two measures were chosen: a measure of pulmonary function, the peak expiratory flow
rate (Cook et al. 1991); and a measure of muscle strength, hand grip strength
(Jacobson-SoUerman and Sperling 1977). Three measurements were taken for
each test and the best score retained for analysis.
F. Self-Rated Health
Self-ratings of health are designed to capture respondents' subjective summary
interpretations of their own medical and functional status and a few of these items
were included. In preliminary analyses, the self-rated overall health status item,
derived from the National Health Interview Survey (NHIS), was related to all
physical as well as mental health measures. Particularly high correlations were
noted with the mobility index (r = .54), the large muscle strength index (r =
.44), the usual and worst pain (r = .43 and r = .42), and with the depressive
symptom index (r = - .43).
G. Hygienic Behaviors
Levels of hygienic behaviors were assessed in HRS respondents for several reasons: they reflect to some extent current functional status (for example, exercise
levels); they may be related to job success, wealth, and retirement status (for
example, alcohol abuse) and health insurance premiums (for example, tobacco
and alcohol use); and they predict future major chronic conditions (for example,
exercise, tobacco use). It should be possible to determine over time the role of
tobacco and alcohol use in productivity and earnings loss as well as health and
retirement status. The cigarette smoking battery was adapted from the NHIS.
Wallace and Herzog
The frequency of current but not past alcohol use was elicited. Four screening
items for alcohol abuse and problem drinking—the "CAGE" items (Ewing and
Rouse 1970)—were also included. Positive responses to any three of four CAGE
questions is the criterion for alcohol dependence of the Diagnostic and Statistical
Manual of Mental Disorders, 3rd edition, revised. Items are phrased in terms of
"ever" use, and will identify the most severe problem drinkers.
Although recreational and leisure activities were of low priority for the HRS
Wave 1, there was a consensus that a few exercise items were needed. Two
items were included assessing light and strenuous physical activity during leisure
pursuits. In order to avoid a white collar/male bias common to surveys which
focus only on recreational exercise, questions on heavy housework and on physical demand levels on the job were added.
Questions eliciting current height and weight were also included, allowing the
calculation of various body mass indices. These self-reported measures are known
to be reasonably accurate (Stevens et al. 1990; Stunkard and Albaum 1981;
Kuskowska et al. 1989), and are typically related to physical performance, various
concurrent medical conditions, and the risk of other conditions and death. An
item permitting the estimation of recent weight change was added as a possible
indicator of changing health status.
IV. Cognitive Health Measures
Cognitive function is likely critical to job performance and retirement behavior. Yet, it has not been explored in previous research and consequently no consensus exists on a brief set of cognitive tests that cover the broad
range of function likely to be important in the retirement transition. The development of cognitive measures presented a major challenge to the design of the
HRS.^ A multidimensional conceptualization of cognitive functioning was
adopted that integrates much of the research on aging and cognition and is spelled
out cogently by Perlmutter (1988). She postulated a three-tiered model, consisting
of the basic mechanisms or processing resources in Tier 1, the knowledge base
in Tier 2, and thinking and strategies in Tier 3. Tier 1 is often referred to as fluid
intelligence and is most closely linked to biological and physical processes; as
such it tends to show the clearest decline in older age. Tier 2, often referred to
as crystallized intelligence, is based on formal education and informal experience
and tends to decline less clearly as people get older. Tier 3 contains strategies
for dealing with knowledge and with one's own cognitive resources and is most
closely related to the concept of metacognition or metamemory. As an adaptive
resource. Tier 3 may continue to grow throughout life. A dementia screening test
2. The HRS Health Working Group had benefitted from the counsel of many talented cognitive psychologists and geriatric psychiatrists. They included Marilyn Albert, John Breitner. Gerda Fillcnbaum, Barry
Fogel. Tamara Harris. Christopher Hertzog. David Hultsch. Judith Kasper, Margie Lachman, Ulman
Lindenberger, William McNaught. Richard Mohs, Marion Perlmutter, John Nesselroade. Timothy Salthouse, and Kathleen Welch. Their assistance is gratefully acknowledged.
S99
SlOO
The Journal of Human Resources
that assesses cognitive impairments was not deemed useful because the prevalence of dementia in 51-61 year olds is extremely low.
A. Memory Test
A test of memory was given top priority for the HRS to assess the ability to
acquire new information and to reasonably represent Tier 1 in Perlmutter's (1988)
conceptualization. Because memory abilities decline as individuals age, this decline might present a disadvantage for the work situation and perhaps affect
retirement timing. The specific HRS measure is a free recall test of 20 nouns,
originally developed for the Established Populations for Epidemiologic Study of
the Elderly in Iowa (O'Hara et al. 1986). Only 2 percent of respondents did not
participate. The distribution of words correctly recalled immediately after the
presentation ranges from 0 to 20 with a mean of 7.6; about 13 percent of participating respondents obtained scores higher than 10. After several intervening survey
questions an average of only 5.5 words could be recalled ("delayed" recall). A
more direct test of speed of processing or Tier 1 abilities such as the Digit Symbol
Substitution Test from the Wechsler Adult Intelligence Scale-Revised was judged
as impossible to administer over the telephone.
B. Abstract Reasoning Test
A test of abstract reasoning was given somewhat lower priority for the HRS than
the memory test but still considered important for the following reasons. First, a
test of abstract reasoning was thought to align itself more with the Tier 2 dimension and thereby to provide the HRS with a comprehensive assessment of cognitive skills. Second, whereas some have suggested that abstract reasoning declines
less clearly with age (Schaie 1983) and thus may not lead as readily to problems
at work, these abilities may become important in compensating for memory deficits noted above. Moreover, decline in abstract reasoning If it does happen, may
be particularly impairing because it robs the individual of the opportunity to
compensate. The specific test chosen for HRS was an abbreviated version of the
Similarities Subtest of the Wechsler Adult Intelligence Scale-Revised. Seven pairs
of words were presented to the respondents with the instruction to describe for
each pair how the two words were alike. Intercoder agreement on scoring of the
open-ended responses was high. The individual pairs of words showed increasing
difficulty. The resulting HRS summary scores distributed approximately normal
with a mean of 6.3 out of a maximum score of 14.
C. Self-Rated Cognitive Functioning
In terms of Tier 3, it was deemed important to assess self-rated cognitive functioning or what some have called metacognition. Two survey questions were written
for the HRS to tap the respondent's global perceptions of his or her quick thinking
and changes in such over the past two years. The term "ability to think quickly"
was intended to describe the concept of cognitive functioning. Six additional
self-report items probing specific aspects of memory functioning were included
Wallace and Herzog
in an experimental module. They were adopted from the Capacity and Change
subscales of the Metamemory in Adulthood questionnaire by Dixon, Hultsch,
and Hertzog (1988).
D. Functioning in Cognitively Demanding Activities of Daily Living (IADLs)
Finally, a few self-report questions about difficulty experienced with cognitive
functioning in everyday life were fashioned after the Personality in Intellectual
Aging Contexts scale by Lachman (Lachman et al., 1982) and also represent
metacognition. The percentages of respondents reporting difficulties are shown
at the bottom of Table 6 and reveal some reported difficulty, particularly for map
and computer use. All. three questions, but particularly the one on computer
use, also yielded a nonnegligible amount of missing data, mostly because many
respondents had no experience with the questioned activity. An index was formed
as the number of these items (excluding the computer item) with which difficulty
was experienced.
If these cognitive measures are valid, they should be related to each other
(concurrent validity) and not related to measures of different concepts (discriminant validity). The two recall measures correlated highly with each other, less
highly but still substantially with the abstract reasoning measure (Table 7). The
correlations between the cognitive performance measures and self-rated ability
to think quickly and the IADL index of cognitive functioning are positive but
small, refiecting similar findings reported in the literature (Herrmann 1983; Morris
1983). Whatever the reasons for the lack of correspondence, the self-report measures were not intended as veridical indicators of cognitive performance, but as
measures of the individual's perception—metamemory—that may well be related
to work and retirement behavior irrespective of the objective performance.
For establishing discriminant validity we examined the relationship between
the cognitive measure and two measures of affective mental health—a measure
of depressive symptoms and a self-report measure of emotional health (Table 7).
The two measures of emotional health correlated strongly with each other (r =
-.57), but they correlated only weakly with the measures of cognitive performance, confirming the discriminant validity of the cognitive as well as of the
emotional measures. Interestingly, the self-rated measures of cognitive functioning and quick thinking correlated at least equally strongly with the emotional
mental health measures as with the cognitive performance measures. Again, similar findings have been reported (O'Hara et al. 1986), suggesting that an individual's perception of his or her cognitive ability is as much colored by emotional
state as by cognitive performance. In terms of construct validity, some study
diseases are related to cognitive functioning, including psychiatric problems,
stroke with remaining health problem, diabetes and heart condition (Table 3).
V. Emotional Health Measures
A. Depressive Symptoms
Despite the likely role of mental illness in job performance and retirement, a
comprehensive assessment of mental illness was not possible because of interview
S101
SI02
The Journal of Human Resources
Table 7
Correlations Between Cognitive and Affective Measures.
HRS Respondents 51-61 years old
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Immediate word recall
(0-20)
Delayed word recall
(0-20)
WAIS-R Similarities
(0-14)
Poor self-rated quick
thinking (1-4)
lADL Difficulty Index
(0-3)
LowCES-D II (1-4)
Poor self-rated emotional health (1-4)
(2)
(3)
(4)
(5)
(6)
(7)
.74**
.33**
-.17**
-.14**
-.14**
.16**
-.17**
.30**
-.15*'
-.14**
.15**
-.15**
- .22*
-.23**
.16**
-.21**
.28**
-.30**
.37**
-.27**
.23**
-.57*
*p < .05. **p < .01. Note; The number of respondents varies between 8.500 and 9.750.
time limitations. Also, the prevalence of most major mental illnesses was thought
to be relatively low, and their detection would not necessarily be informative
with the current study design. It was deemed that affective function, as indicated
by depressive symptom presence and frequency, was most important to assess
because of its relatively high prevalence and known relation to physical health
and functioning. The measure that was chosen for the HRS is the Center for
Epidemiological Studies Depression (CES-D) Scale (Radloff 1977). This scale has
been widely used with young as well as elderly respondents. Although the full
scale has 20 items, we used a short version developed for the Established Populations for Epidemiologic Study of the Elderly (Kohout et al. 1993).
An exploratory factor analysis of the HRS data revealed factors of affect,
psychosomatic symptoms, and interpersonal problems, confirming previous factor analytic findings for the CES-D. The coefficient alpha for all eleven items was
.84, indicating high internal consistency. A summary score formed as the average
of all 11 items yielded a skewed distribution, with a mean of 3.6 out of a maximum
score of 4.0.
To explore concurrent validity, one would expect the CES-D measure to be
related to other measures of emotional health. The relationships of the CES-D
summary score with the self-rating of emotional health written for the HRS (already presented in Table 7) and with the self-reported presence of "emotional,
nervous, or psychiatric problems" (Table 3) are substantial. Both relationships
support the concurrent validity of the CES-D as well as of the two other selfreports of emotional problems. The relationships with the cognitive performance
measures are low, supporting the CES-D's discriminant validity. If the CES-D
Wallace and Herzog
measure had construct validity, we would expect it to be related to physical
health problems that depress people who are afflicted. Depressive symptoms are
more frequent among persons suffering from all severe chronic conditions, but
particularly from stroke, chronic lung disease and heart condition (Table 3).
The CES-D measure also was strongly related to the vitality measure discussed
earlier. A factor analysis of all the items from the two measures indicated that
the vitality items were most closely linked to the psychosomatic dimension of the
depression measure. This suggests that the two concepts—though theoretically
distinct—are not empirically distinguishable in the HRS and therefore that discriminant validity cannot be established. How much this is a function of conceptual similarity or of measurement similarity cannot be determined with these HRS
data.
B. Subjective Well-Being
The subjective quality of life is a broad concept that has been studied from a
number of perspectives and under a variety of labels. The most useful perspective
for the HRS is one that conceptualizes overall subjective quality of life as a
composite of the quality of life in different arenas such as work but also family
life, leisure, economic situation, and health. There is now substantial evidence
that satisfaction with life as a whole can be appropriately modelled as a simple
linear function of satisfactions experienced in the various domains (Andrews and
Withey 1976). As people age, their adjustment to changing roles and competencies
is expected to be influenced by the potential for satisfactions they perceive in the
different arenas of life. Thus, the retirement decision is most likely influenced by
the satisfaction derived from work and the satisfactions experienced in alternative
domains of life such as leisure or family life. It is for these reasons that none of
the various multiple-item scales that have been developed to assess overall life
satisfaction was adopted but a number of domains were chosen, including health,
financial situation, family life, and others, and satisfaction with each was measured.
VI. Health Care Service Utilization
There were several reasons for assessing various types of health
service utilization on the part of HRS participants. First, health care utilization
is an indicator of the burden of disease and disability. Second, it can provide
construct validity for the health measures. Third, it may have a direct bearing on
the likelihood of staying in the labor force. Fourth, through associated out-ofpocket costs, it may impact on economic status.
The total number of doctor visits is a widely used indicator of utilization and
was awarded a high priority in HRS. The National Health Interview Survey
utilization question was modified to a one year time frame and extended to all
kinds of ambulatory care settings (for example, emergency rooms). The HRS also
added a question about professional nursing care in the home and one about time
spent on self-care. Hospitalization information included the number of admissions
SI03
S104
The Journal of Human Resources
over the previous year and the total number of nights stayed. The same information was collected about nursing home stays.
Strong consideration was given to surveying for current prescription and overthe-counter medications. It was recognized that medications are indicators of
access to medical services and presence of important chronic illnesses, and that
they represent some of the cost burden of health care. Additionally, several types
of medications might have direct detrimental effects on work performance. Unfortunately, an elaborate and time consuming question strategy is required to ascertain the drugs taken, dosages and adherence to prescribed regimens, and this was
abandoned.
Preliminary findings support the construct validity of the HRS health care utilization measures. Information in Table 4 indicates that persons with severe heart
conditions, with cancer undergoing treatment, and with stroke with remaining
health consequences reported the highest number of nights in the hospital of all
currently active medical conditions investigated in the HRS. Cancer patientstogether with mental and heart condition patients—are also the ones reporting
the highest number of doctor visits. On the other hand, stroke sufferers and
persons with a severe lung condition require the highest number of sick days,
persons with a severe heart condition the most self care. Although some of these
effects are confounded by severity being defined partly in terms of health care
utilization (see Table 1), this cannot explain the entirety of these effects.
VII. Summary and Discussion
The HRS contains a broad range of health measures including
measures of physical, cognitive, and emotional health. Global self-assessments,
specific function reports, defined medical conditions, symptoms, hygienic behaviors, health care utilization and, albeit in a somewhat more experimental mode,
performance measures were used to explore health status. This will permit the
analyst to gauge the general significance of health for work and retirement as well
as probe specific aspects. It will also permit contrasting more subjective with
more objective health measures. Finally, it will permit describing illness progression from risk factors to disease conditions to impairments to institutionalization
and finally to death; examining the impact of this progression on work and retirement patterns; and to a more limited degree, assessing illness progression according to occupation. To summarize, health is obviously a major factor affecting
retirement timing. It is also clear that health is related to socioeconomic factors.
Therefore a good measure of health is needed to properly model retirement behavior and to avoid biasing the estimates of socioeconomic factors.
We are generally pleased with the quality of the HRS health measures as revealed in the necessarily preliminary analyses reported in this paper. Measures
showed concurrent, discriminant, and construct validity where we were able to
evaluate them. The response distributions were typically encouraging, resembling
data collected with similar measures in other studies and, where such previous
information was not available, approximating sensible distributions. More work
remains, however, in building a case for the validity of the HRS health measures.
Wallace and Herzog
There are several limitations to the health measurement approach taken in HRS
that should be noted. While most items used have been shown to be at least
moderately accurate in other studies, there will be little external validation as
part of the HRS protocol. Resource limitations precluded acquisition of health
records for this purpose. Also, the selection of some items that are not represented at all or at least not verbatim in other national surveys lessens the ability
to compare and contrast HRS findings with these studies. In addition, for many
dimensions of the health history, there is by design no attempt to obtain lifelong
events or characterize long-term changes in various health states. This limits the
ability to determine whether the duration of various conditions relates to the
social or economic variables or outcomes. Also, it was not possible to survey for
all medical conditions, symptoms, and dysfunctions, and only the most common
ones, of public health importance, were queried. However, an open-ended item
allows respondents to offer other health conditions and problems of note. Finally,
multi-item scales such as the CES-D or the CAGE instrument were considerably
reduced in length.
Other limitations relate to subsequent waves of the HRS. The second wave,
performed in 1993-94, does not generally attempt to ascertain the fluctuations in
health or functional status in the two year interval between survey waves. Rather
it defines health status mostly at the time of each survey, and may miss some
variation in health status. Also, while many of the health items have been used
in telephone surveys, whether there will be mode effects in response accuracy is
uncertain.
In summary, despite restrictions in time and resources and occasional controversy over the best measures of health and cognitive phenomena, we believe that
the baseline HRS survey contains a comprehensive set of measures of health
and function. Not only have these measures proven to relate to each other in
theoretically sensible ways, but they will hopefully provide important predictors
to explore the role of health factors in the retirement decision and post-retirement
health and economic events.
References
Anderson, K. H., and R. V. Burkhauser. 1984. "The Importance of the Measure of
Health in Empirical Estimates of the Labor Supply of Older Men." Economic.^ Letters
16:375-80.
. 1985. "The Retirement-Health Nexus: A New Measure of an Old Puzzle."
Journal of Human Resources 20(3):315-30.
Andrews, F. M., and S. B. Withey. 1976. Social Indicators of Well-Being: Americans'
Perceptions of Life Quality. New York: Plenum Press.
Bartel, A., and P. Taubman. 1979. "Health and Labor Market Success: the Role of
Various Diseases." Review of Economics and Statistics 61:1-8.
Bergner, M., R. A. Bobbitt, W. B. Carter, and B. S. Gilson. 1981. "The Sickness Impact
Profile: Development and Final Version of a Health Status Measure." Medical Care
19(«): 787-805.
Bound, J. 1991. "Self-Reported Objective Measures of Health in Retirement Models."
Journal of Human Resources 26(l):106-38.
SI05
SI06
The Journal of Human Resources
Burkhauser, R. V., J. S. Butler, and J. M. Mitchell. 1986. "Effects of Arthritis on Wage
Earnings." Journal of Gerontology 41:277-81.
Chirikos, T. N. 1993. "The Relationship Between Health and Labor Market Status."
Annual Review of Public Health 14:293-312.
Cook, N. R., D. A. Evans, P. A. Scherr, F. E. Speizer, J. O. Taylor, and C. H.
Hennekens. 1991. "Peak Expiratory Flow Rate and 5-Year Mortality in an Elderly
Population." American Journal of Epidemiology 133:784-94.
Cote, R. A., ed. 1982. Systematized Nomenclature of Medicine. Skokie, IL: College of
Pathologists.
Dixon, R. A., D. F. Hultsch, and C. Hertzog. 1988. "The Metamemory in Adulthood
(MIA) Questionnaire. Psychopharmacology Bulletin 24:671-88.
Ewing, J. A., and R. A. Rouse. 1970. "Identifying the 'Hidden Alcoholic.'" Presented at
the 29th International Congress on Alcohol and Drug Dependence, Sydney, Australia.
Ferraro, K. F. 1980. "Self-Ratings of Health Among the Old and the Old-Old. Journal of
Health and Social Behavior 21:377-83.
Guralnik, J., and A. Z. LaCroix. 1992. "Assessing Physical Function in Older
Populations." In The Epidemiologic Study ofthe Elderly, ed. R. Wallace and R.
Woolson. New York: Oxford University Press.
Hennessey, J. C , and J. M. Dykacz. 1989. "Projected Outcomes and Length of Time in
the Disability Insurance Program." Social Security Bulletin 52:2-41.
Herrmann, D. J. 1983. "Questionnaires about Memory." In Everyday Memory: Actions
and Absent-Mindedness, ed. J. E. Harris and P. E. Morris. London: Academic Press.
Huntley, J. C , D. Brock, A. Ostfeld, J. Taylor, R. Wallace, and M. Lafferty. 1986.
"Established Populations for Epidemiologic Study ofthe Elderly." Resource Data
Book. National Institute on Aging. NIH Publication No. 86-2443.
Jacobson-Sollerman, C , and L. Sperling. 1977. "Grip Function in the Health Hand in a
Standardized Hand Function Test." Scandinavian Journal of Rehabilitative Medicine
9:123-29.
Kohout, F. J., L. F. Berkman, D. A. Evans, and J. Cornoni-Huntley. 1993. "Two
Shorter Forms ofthe CES-D Depression Symptoms Index." Journal of Aging and
Health 5:179-93.
Kuskowska, A., P. Karlsson, M. Stolt, and S. Rossner. 1989. "The Predictive Validity
of Body Mass Index Based on Self-Reported Weight and Height." International
Journal of Obesity 13:441-53.
La Rue, A., L. Bank, L. Jarvik, and M. Hetland. 1979. "Health in Old Age: How Do
Physicians' Ratings and Self-Ratings Compare?" Journal of Gerontology 34:687-91.
Lachman, M. E., P. B. Baltes, J. R. Nesselroade, and S. L. Willis. 1982. "Examination
of Personality-Ability Relationships in the Elderly: The Role of the Contextual
(interface) Assessment Mode." Journal of Research in Personality 164:85-501.
Maddox, G, and E. Douglas. 1973. "Self-Assessment of Health: A Longitudinal Study of
Elderly Subjects." Journal of Health and Social Behavior 14:87-93.
McHorney, C. A., J. E. Ware, Jr., and A. E. Raczek. 1993. "The MOS 36-ltem
Short-Form Health Survey (SF-36): II. Psychometric and Clinical Tests of Validity in
Measuring Physical and Mental Health Constructs." Medical Care 31:247-63.
Mitchell, J. M., and K. H. Anderson. 1989. "Mental Health and the Labor Force
Participation of Older Workers.'' Inquiry 26:262-71.
Mobily, P. R., K. A. Herr, M. K. Clark, and R. B. Wallace. 1994. "An Epidemiologic
Assessment on Pain in the Elderly." Journal of Aging and Health 6:139-54.
Morris, P. E. 1983. "The Validity of Subjective Reports on Memory." In Everyday
Memory: Actions and Absent-Mindedness, ed. J. E. Harris and P. E. Morris. London:
Academic Press.
Wallace and Herzog
Myers, R. J. 1982. "Why Do People Retire from Work Early?" Aging and Work
5(2):83-91.
Nagi, S. Z. 1965. "Some Conceptual Issues in Disability and Rehabilitation." In
Sociology and Rehabilitation, ed. M. B. Susman. Washington, D.C: American
Sociological Association.
O'Hara, M. W., J. V. Hinrichs, F. J. Kohout, R. B. Wallace, and J. H. Lemke. 1986.
"Memory Complaints and Memory Performance in the Depressed Elderly."
Psychology and Aging 1:208-14.
Patrick, D. L., and P. Erickson. 1993. Health Status and Health Policy: Quality of Life
in Health Care Evaluation and Resource Allocation. New York: Oxford University
Press.
Perlmutter, M. 1988. "Cognitive Potential Throughout Life." In Emergent Theories of
Aging, ed. J. E. Birren and V. L. Bengtson. New York: Springer Publishing Co.
Pope, A. M., and Tarlov, A. R., eds. 1991. Disabilities in America: Toward a National
Agenda for Prevention. Washington, D. C : National Academy Press.
Radloff, L. S. 1977. "The CES-D Scale: A Self-Report Depression Scale for Research in
the General Population." Applied Psychological Measurement 1:385-401.
Quinn, J. F., and Burkhauser, R. V. 1990. "Work and Retirement." In Handbook of
Aging and the Social Sciences, Third Edition, ed. R. H. Binstock and L. K. George.
New York: Academic Press.
Schaie, K. W. 1983. "The Seattle Longitudinal Study: A Twenty-One Year Exploration
of Psychometric Intelligence in Adulthood." In Longitudinal Studies of Adult
Psychological Development, ed. K. W. Schaie. New York: Guilford Press.
Stevens, J., J. E. Kail, R. Waid, and P. C. Gazes. 1990. "Accuracy of Current 4-year
and 28-year Self-Reported Body Weight in an Elderly Population." American Journal
of Epidemiology 132:1156-63.
Stewart, A. L., and J. E. Ware, eds. 1992. Measuring Functioning and Well-Being.
Durham: Duke University Press.
Stunkard, A. J., and J. M. Albaum. 1981. "The Accuracy of Self-Reported Weights."
American Journal of Clinical Nutrition 34:1593-99.
U.S. Bureau of the Census. 1994. Statistical Abstract of the United States: 1994 (114th
edition). Washington, D.C: GPO.
Williams, S. J. 1994. "Measuring Functional Limitations and Sickness Impact: A Critical
Review of the FLP and SIP." In Measuring Health and Medical Outcomes, ed. C
Jenkinson. London: UCL Press.
Wright, L. 1994. "The Long and Short of It: The Development of the SF-36 General
Health Survey." In Measuring Health and Medical Outcomes, ed. C Jenkinson.
London: UCL Press.
SI07