Close
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Processes of loss and recovery of physical abilities in a longitudinal study of Americans 70 years of age and older
(USC Thesis Other)
Processes of loss and recovery of physical abilities in a longitudinal study of Americans 70 years of age and older
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI
films the text directly from the original or copy submitted. Thus, some
thesis and dissertation copies are in typewriter face, while others may
be from any type of computer printer.
The quality of this reproduction is dependent upon the quality of the
copy submitted. Broken or indistinct print, colored or poor quality
illustrations and photographs, print bleedthrough, substandard margins,
and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete
manuscript and there are missing pages, these will be noted. Also, if
unauthorized copyright material had to be removed, a note will indicate
the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by
sectioning the original, beginning at the upper left-hand comer and
continuing from left to right in equal sections with small overlaps. Each
original is also photographed in one exposure and is included in
reduced form at the back of the book.
Photographs included in the original manuscript have been reproduced
xerographically in this copy. Higher quality 6 " x 9 " black and white
photographic prints are available for any photographs or illustrations
appearing in this copy for an additional charge. Contact UMI directly
to order.
A Bell & Howell Information Company
300 North Zeeb Road. Ann Arbor. M l 48106-1346 USA
313/761-4700 800/521-0600
PROCESSES OF LOSS AND RECOVERY OF PHYSICAL ABILITIES
IN A LONGITUDINAL STUDY OF AMERICANS
70 YEARS OF AGE AND OLDER
by
Adrienne Heidt Pixley Mihelic
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Gerontology)
December 1995
Copyright 1995 Adrienne Heidt Pixley Mihelic
UMI Number: 9625266
Copyright 1995 by
Mihelic, Adrienne Heidt Pixley
All rights reserved.
UMI Microform 9625266
Copyright 1996, by UMI Company. All rights reserved.
This microform edition is protected against unauthorized
copying under Title 17, United States Code.
UMI
300 North Zeeb Road
Ann Arbor, MI 48103
UNIVERSITY OF SOUTHERN CALIFORNIA
THE GRADUATE SCHOOL
UNIVERSITY PARK
LOS ANGELES, CALIFORNIA 90007
This dissertation, written by
Adrienne Heidt Pixley Mihelic
under the direction of /lsk Dissertation
Committee, and approved by all its members,
has been presented to and accepted by The
Graduate School, in partial fulfillment of re
quirements for the degree of
DOCTOR OF PHILOSOPHY
Dean o f Graduate Studies
Date
DISSERTATION COMMITTEE
, %
.....
Chairperson
This work is dedicated to two extraordinary women, my grandmothers.
Margaret Steele Moore Walton
Marie Angel Gabrielle Jova Pixley
ii
ACKNOWLEDGEMENTS
The first round of thanks goes to my dissertation committee, Prof. Eileen
M. Crimmins, Prof. Vern L. Bengtson, Prof. David Heer, Prof. Merril
Silverstein, and Prof. Lisa Greenwell. As a committee, they were patient,
encouraging, and challenging; individually each has made an important mark on
my progress to this point.
Vern taught me innumerable little tricks to improve my communication
skills that have made this opus considerably more readable than it otherwise would
have been. I like to think that this has ultimately helped me to organize my
thoughts more clearly, and thus to think better. Vern is the director of the
National Institute on Aging Multidisciplinary Research Training Grant in
Gerontology at the University of Southern California, which funded my first years
as a doctoral student. Vern used this program to enrich the academic lives of the
trainees with weekly seminars, critical debates, and professional socialization.
And, it is thanks to the training grant that I have been able to work with Eileen
Crimmins as my preceptor.
Sometimes we do good things for irrational reasons. I am not sure why I
decided to take David Heer’s course, "Methods of Population and Ecological
Analysis." It met at eight o’clock in the morning, in an icy classroom, four days a
week, during the summer. I had taken a demography of aging class the semester
before, but this was my first exposure to the nitty-gritty. One does not complete
this class without becoming highly sensitized to details, and that has been a great
help to me. Prof. Heer brought this same careful approach to reading both my
master’s thesis and this work.
Merril can always be counted on to help think about an issue, and
invariably he sees several more sides of an issue than I. In this same spirit he
suffered with me through weeks of entirely incomprehensible error messages from
an undocumented software program I was trying. Lisa gave me the extra push on
a number of occasions to be persistent in my attention to details. What this means
is that when I thought I was at a dead-end trying to figure out the right way to
address an issue, I would consult with Lisa (expecting her to confirm my position).
Her response was invariably to recommend some other resources and then to
monitor my progress. Both Lisa and Merril must be thanked for always acting like
the tedious tasks are fun.
Writing a dissertation has its ambiguous moments when one does not quite
know what to think or how to feel about the process and most especially about
one’s latest effort. Left to my own devices, I tend to think that I have produced a
final product. Invariably Eileen returned my offerings with a list of insightful
suggestions for improvements, typically accompanied by the comment that I
"should feel encouraged." Luckily I am vulnerable to the power of suggestion,
and consequently, feelings of devastation were generally averted by this hint.
Eileen’s mentorage has really gone far beyond marshalling a good attitude.
She coached me over every hurdle and taught me to negotiate the sometimes
obscure terrain of academia; it is difficult to think of any way in which her good
counsel has not smoothed my path. Working with her as student, teaching
assistant, research assistant and trainee has been more than instructive, it has been
delightful. There are not good enough words to thank her for all that she has done
for me.
It is an understatement to say that without my parents, Francois and Eleanor
Pixley, I would not be here. This is more than a biological necessity. They
socialized me to think that I could do anything; I have not always thanked them for
this. They are a responsible team, though, and having gotten me in a precarious
position, they never let fear of failure make much of an inroads on my progress.
My mother has had a profound influence on how I see the world, and therefore,
indirectly, on the course of study I have chosen. I must have inherited my work
habits from my father, but he has given me an equal enthusiasm for fun. My love
of learning is a reflection of their own inquiring minds and their investment in my
growth from the very beginning — opening a Montessori school for me in our
backyard.
If my parents are responsible for my existence, my husband, Mark, is
responsible for my survival, and for my surviving with joy. He kept me from
being a "victim" of graduate school. Thanks to him, I was never hungry, cold,
nor without love and words of encouragement. He made me laugh when I was
nervous, and play as a way of life.
v
Federally funded training programs and small grant opportunities have done
much to encourage students to pursue careers in much needed areas of research.
These programs have enabled many of us to receive superior training and to work
with leaders in our fields while minimizing the financial trauma graduate school
often entails. Preparation for this work was supported by the Multidisciplinary
Research Training Grant in Gerontology, National Institute on Aging No.
5T32AG00037. This research was funded by a dissertation grant from the Agency
for Health Care Policy and Research No: R 03 HS08459.
TABLE OF CONTENTS
PAGE
ACKNOWLEDGEMENTS .................................................................................... iii
LIST OF FIGURES ............................................................................................. xiii
LIST OF TABLES ................................................................................................. xiv
ABSTRACT ............................................................................................................ xviii
CHAPTER I
INTRODUCTION ..................................................................................... 1
A. Research Questions ...................................................................... 4
B. Significance .................................................................................... 7
C. Overview of the Chapters .......................................................... 8
1. Conceptual Framework .................................................. 9
2. Data and Sample Dynamics ......................................... 9
3. Evaluating the Effects of Nonresponse
in the LSOA .................................................................... 10
4. Using Items from the ADL and IADL Scales
to Measure Change in Functioning ........................... 11
5. Patterns of Change in Functional Health .................. 12
6. Processes of Change in Functional Health ............... 13
7. Discussion .......................................................................... 15
CHAPTER II
CONCEPTUAL FRAMEWORK ......................................................... 16
A. Introduction ................................................................................... 16
B. The Outcome: Physical Functioning ...................................... 17
1. Conceptualizing Physical Functioning ....................... 17
2. Approaches to Representing Changes
in Functioning.................................................................... 24
TABLE OF CONTENTS (Continued)
C. Predictors of Change in Functioning ...................................... 27
1. Demographic, Economic,
and Social Characteristics ..................................................... 29
2. Health Behaviors .................................................................... 34
3. Diseases and Conditions ...................................................... 35
D. Summary and Elaboration:
Introduction to Alternative Conceptualizations .................... 38
1. Summary of Hypothesized Predictors ........................ 39
2. Reconceptualizing the Representation
of Change in Functioning: An Introduction ............ 42
CHAPTER III
DATA AND SAMPLE DYNAMICS .................................................. 49
A. Introduction .................................................................................... 49
B. Data .................................................................................................. 49
1. Longitudinal Design ....................................................... 51
2. Linked Data ...................................................................... 52
3. Distribution of Demographic Characteristics .......... 53
C. Sample Dynamics .......................................................................... 54
1. Introduction ...................................................................... 54
2. Sample Dynamics ............................................................ 54
D. Discussion ....................................................................................... 58
CHAPTER IV
EVALUATION OF THE EFFECTS OF NONRESPONSE
OVER FOUR WAVES OF THE LSOA ............................................ 65
A. Introduction .................................................................................... 65
TABLE OF CONTENTS (Continued)
B. Background ..................................................................................... 66
1. Individual Characteristics
Affecting Nonresponse .................................................. 67
2. Effects of Survey Design and
Sampling Characteristics ............................................. 69
3. Summary ............................................................................ 70
C. Modeling Loss to Follow-up ....................................................... 72
1. Analytic Approach ........................................................... 73
2. Model and Measures ....................................................... 75
3. Results of Multivariate Analysis
of Loss to Follow-up ..................................................... 76
4. Summary ............................................................................ 80
D. Adjusting for Loss to Follow-up ............................................... 82
1. Background: Correcting for Nonresponse ................. 82
2. Effect of Nonresponse A djustm ents............................. 84
3. Analytic Approach to Compensating
for Nonresponse .............................................................. 86
E. Summary and Discussion ............................................................. 90
CHAPTER V
USING ITEMS FROM ADL AND IADL SCALES
TO MEASURE CHANGE IN FUNCTIONING .............................. 98
A. Introduction .................................................................................... 98
1. The Issues: An Introduction ......................................... 100
2. Description of Self-report Functioning
Items Available in the LSOA ...................................... 102
3. Organization of the Chapter............................................ 102
B. Competing Approaches to Combining Items from
ADL and IADL Scales................................................................. 103
ix
TABLE OF CONTENTS (Continued)
1. Introduction ...................................................................... 103
2. Dimensionality ................................................................ 104
3. Scaling Properties of ADL
and IADL Items................................................................. 112
C. Implication of Question Wording ............................................. 117
1. Response Bias .................................................................. 118
2. Evaluation of Impact on Question Type on the
Representation of Change in Functioning ................. 121
D. Discussion ....................................................................................... 124
CHAPTER VI
PATTERNS OF CHANGE IN FUNCTIONAL HEALTH ........... 145
A. Introduction .................................................................................... 145
B. Background .................................................................................... 147
C. Sample ............................................................................................. 148
D. Representing Change in Functional Health .............................. 149
1. Interpreting the Categories of Health Status ............. 150
2. Patterns of Change in Level of Functioning ........... 151
E. Multivariate Analysis .................................................................. 154
1. Dependent Variable ....................................................... 155
2. Independent Variables .................................................... 156
3. Analysis of Declines ...................................................... 160
4. Analysis of Fluctuations ................................................ 163
5. Factors Affecting Probability
of Decline and Fluctuation .......................................... 164
F. Discussion ....................................................................................... 165
x
TABLE OF CONTENTS (Continued)
CHAPTER VII
PROCESSES OF CHANGE IN
FUNCTIONAL HEALTH ...................................................................... 182
A. Introduction ..................................................................................... 182
B. Test of a Simple Markov Model ............................................... 184
C. Multivariate Analysis of Specific Transitions ........................ 186
1. Sample ................................................................................ 186
2. Measures ........................................................................... 187
3. Analysis .............................................................................. 189
4. Predictors of Transitions to
Worse Functioning by Initial Status .......................... 190
5. Predictors of Transitions to
Better Functioning by Initial Status ........................... 195
D. Multivariate Analysis of Types of Transitions ..................... 196
1. Predictors of Decline Over an Interval ..................... 198
2. Predictors of Improvement Over an Interval .......... 199
E. Summary and Comparison of Models ...................................... 200
1. Models Predicting Decline ........................................... 200
2. Models Predicting Improvement ................................. 201
F. Discussion ........................................................................................ 202
1. Differences in Transitions
by Level of Change ....................................................... 202
2. Effect of Previous Experience
on Subsequent Change .................................................. 204
3. Conclusion ........................................................................ 207
xi
TABLE OF CONTENTS (Continued)
CHAPTER VIII
SUMMARY AND DISCUSSION ........................................................ 215
A. Summary and Discussion of Findings .......................................... 216
1. Results of Incorporating Information
from Multiple Observations ......................................... 216
2. Summary of Findings Regarding Predictors
of Functioning Change .................................................. 220
B. Significance of Findings and Implications
for Future Research .......................................................................... 229
REFERENCES ........................................................................................................ 238
xii
LIST OF FIGURES
PAGE
Figure II-l World Health Organization Model
of Disease Consequences ........................................................... 46
Figure II-2 Verbrugge’s Conceptualization of Intrinsic
and Extrinsic Ability, Embellished ......................................... 47
Figure II-3 Synthesis and Elaboration of Theoretical
Models Implicit in the Literature ............................................ 48
Figure III-l Transitions in Respondent Status
Over 4 Waves in the LSOA ..................................................... 61
Figure V -l Transition Rates: Loss of IADL — Three Measures ........... 131
Figure V-2 Transition Rates: Loss of ADL — Three Measures ............. 132
Figure VI-1 Possible Transitions Between Levels of Functioning .......... 171
Figure V II-1 Representation of Selected Processes of Change ................. 209
LIST OF TABLES
PAGE
Table III-l Descriptive Statistics for Nonresponse
in 4 Waves of the LSOA ........................................................... 62
Table III-2 Comparison of Characteristics at Baseline
for the Total Sample and for the Sample Selected
to Include Only Those Who Respond
at Every Wave ............................................................................. 63
Table III-3 Response Patterns for Three Waves of Follow-up ............... 64
Table IV -1 Variable Definitions and Coding
by Observation Plan .................................................................... 94
Table IV-2 Odds Ratios for Significant Predictors
of Nonresponse ............................................................................. 95
Table IV-3 Predicted Mean Probability of Becoming
a Nonrespondent .......................................................................... 96
Table IV-4 Comparison of Frequencies for ADLs and IADLs
in Each Wave Without Weights, With Only LSOA
Sample Correction Weights, and with Weights
Constructed to Correct for Probability
of Nonresponse ................................................................................... 97
Table V -l Exploratory Factor Analysis with 12 Measures
of Functioning Difficulty from ADL & IADL Scales
at 4 Interviews; Standard Regression Coefficients for
Oblique Rotation, Rounded and Multiplied by 100 ............ 133
Table V-2 Exploratory Factor Analysis with 12 Measures
of Functioning Disability from ADL & IADL Scales
at 4 Interviews; Standard Regression Coefficients for
Oblique Rotation, Rounded and Multiplied by 100 .................. 134
xiv
LIST OF TABLES (Continued)
Table V-3 Exploratory Factor Analysis with 12 Measures
of Receipt of Help from ADL & IADL Scales
at 4 Interviews; Standard Regression Coefficients for
Oblique Rotation, Rounded and Multiplied by 100 ............ 135
Table V-4 Exploratory Factor Analysis with 12 Measures .
of Change in Functioning Difficulty from ADL & IADL
Scales at 4 Interviews; Standard Regression
Coefficients for Oblique Rotation,
Rounded and Multiplied by 100 ............................................... 136
Table V-5 Exploratory Factor Analysis with 12 Measures of Change
in Functioning Ability from ADL & IADL Scales
at 4 Interviews; Standard Regression Coefficients for
Oblique Rotation, Rounded and Multiplied by 100 ............ 137
Table V-6 Exploratory Factor Analysis with 12 Measures of Change
in Receipt of Help from ADL & IADL Scales at
4 Interviews; Standard Regression Coefficients for
Oblique Rotation, Rounded and Multiplied by 1 0 0 .............. 138
Table V-7 Coefficients of Reproducibility for Alternate Orderings
of Have Difficulty Measures of Six IADL Functions
at Four Interviews ....................................................................... 139
Table V-8 Coefficients of Reproducibility for Alternate
Orderings of Disability Measures of Six IADL Functions
at Four Interviews ............................................................................. 140
Table V-9 Coefficients of Reproducibility for Alternate Orderings
of Receipt of Help Measures of Six IADL Functions
at Four Interviews ............................................................................. 141
Table V-10 Proportions Declining for Individual Functions
Across 3 Intervals ....................................................................... 142
Table V - ll Proportions Improving for Individual Functions
Across 3 Intervals ............................................................................. 143
xv
LIST OF TABLES (Continued)
Table V-12 Relationship of ADLs to IADLs ................................................ 144
Table VI-1 Sample Characteristics at Baseline Interview ........................ 172
Table VI-2 Distribution of Functioning State
at Four Interviews ....................................................................... 173
Table VI-3 Frequency and Percent Distribution of Patterns
of Change in Functioning Health Over Four Years ........... 174
Table VI-4 Percentage Distribution by Sex of Types of Change
in Functioning State .................................................................... 175
Table VI-5 Percentage Improving, Declining, Fluctuating
and Staying at the Same Level of Functioning for
Each of Three Intervals and Over the Three Intervals ...... 176
Table VI-6 Patterns of Change and Functioning Status ........................... 177
Table VI-7 Effects of Sociodemographic, Economic, Health
and Health Care Utilization Characteristics
on the Conditional Odds of Having a Pattern
of Change in Functioning Characterized as
Decline or Fluctuation ................................................................ 178
Table VI-8 Effects of Sociodemographic, Economic, Health
and Health Care Utilization Characteristics on the
Conditional Odds of Having a Pattern of Change in
Functioning Characterized as Decline, by Sex .................... 179
Table VI-9 Effects of Sociodemographic, Economic, Health
and Health Care Utilization Characteristics on the
Conditional Odds of Having a Pattern of Change in
Functioning Characterized as Fluctuation, by Sex .............. 180
Table VI-10 Predicted Mean Probability of Declining or Fluctuating
for Specific Values of Significant Independent Variables
with All Other Characteristics as Observed .......................... 181
xvi
LIST OF TABLES (Continued)
Table VII-1 Percent in Each Initial Status Making a Transition
to Each Final Status for Every Interval ................................. 210
Table VII-2 Predicted and Actual Prevalence of Disability ..................... 211
Table VII-3 Odds Ratios for Significant Predictors of Transitions
to Worse Health by Initial Status ............................................ 212
Table VII-4 Odds Ratios for Significant Predictors of Transitions
to Improved Health by Initial Status ...................................... 213
Table VII-5 Odds Ratios for Significant Predictors
of Generalized Transitions ........................................................ 214
Table V III-1 Comparison of Odds Ratios Predicting Decline
Over an Interval with Patterns of Decline ............................. 236
Table VIII-2 Summary of Patterns of Significance
for All Models of Decline ......................................................... 237
xvii
ABSTRACT
Loss and recovery of physical abilities can occur throughout older age, and
involve both gradual and relatively sudden change. In this study, four observations
of a panel of persons 70 years of age and older provide more explicit detail on
how characteristics associated with change in functioning affect ability than has
previously been reported from this type of data. A combination of longitudinal
approaches reveals: (a) those who had previously experienced some functioning
loss (even when ability is fully recovered) are more vulnerable to subsequent loss
of functioning than those who have not had previous disability; (b) the number of
physical limitations a person has can be used as an indicator of risk of further
decline in self-care and independent living abilities; (c) diseases and other health
conditions vary in their effects on functioning — some are likely to result in
declines and debilitation over the six-year period, while others tend to be
associated with shorter periods of disability because of recovery or adaptation; (d)
while the effects of socioeconomic and demographic characteristics are largely
subsumed under the more immediate health characteristics included in the analyses,
there are clear differences in propensity to gain or lose ability among population
subgroups, and these warrant further investigation.
I. INTRODUCTION
Loss and recovery of ability to maintain independent living and perform
basic personal care at older ages is a complex and dynamic process. This study
uses four waves of panel data from a sample o f older Americans to develop a more
complete description of the health experience at older ages than has been available
heretofore. This approach differs from previous research in using multiple
observations to construct sequences of functioning status for individuals over a six-
year period. Representing change in this way provides new insights about
movement into and out of disability. Individual histories o f functioning status
depict the diversity o f experience among older persons as they age, and knowing
this history helps predict future changes in ability. A two-pronged approach to
representing change in ability over time yields new information about how several
major diseases and conditions affect the risk of functional decline.
Change in functional health is important because of its implication for older
persons’ independence, and because ability to perform basic tasks is often used as a
screen for program eligibility and service provision. A better understanding of the
processes of functioning health change is needed to identify appropriate
interventions, to determine which individuals are most likely to benefit from
interventions, and to establish the ideal timing for interventions. Appropriately
timed and targeted preventions and interventions offer the possibility of reducing
time spent with disabilities before death.
1
There is considerable diversity in characteristics and experience among
older persons, and this is certainly true of health experiences. Understanding the
variety of patterns of health change an individual might have, or better, having a
sense of the trajectory an individual with particular characteristics is likely to take,
may provide important information for the planning and provision of services.
There is a gap, however, between the functioning experience an individual
has and the way this experience is typically represented in research analyses. Our
own experiences and those of older friends and relatives suggest that under certain
circumstances, a person who has lost some functioning ability is able to "bounce
back," but under other circumstances, a loss of function may be the harbinger of
further declines.
Despite this common-sense understanding of the diversity of health changes
with age, research has consistently represented this extremely dynamic and variable
process in very simple aggregate terms. Experience is represented as change
between two points in time, and usually between two states (i.e. disability and no
disability), described as improvement, decline, or maintenance of functioning
status. This approach has formed something of a paradigm for the way we think
about health change; and, although representing change in a brief time slice that
can be characterized as improvement, decline, or stability in status is a convenient
and reasonable summary of what a person experiences between any two points in
time, it overly simplifies what actually happens to people over a number of time
points. One problem with having such a gap between the experience we are trying
to describe and explain and the way this experience is represented in analytic
studies is that the resulting descriptions and predictions may be inaccurate.
The goals of this research are (1) to describe empirically, for a large
representative sample of older Americans, the diversity of experience in
functioning change over multiple observation points; (2) to introduce into a
commonly-used analytic model of health change a simple method for incorporating
more information on the history of functioning change, in order to test the
hypothesis that including this information on prior status will improve the
prediction of subsequent health change; and (3) to identify salient predictors of
change in functioning. The practical importance of these goals is to steer our
conceptualization of health change away from overly simplistic representations
toward a more realistic approach that incorporates more information and more
individual variability.
This study elaborates the way research on change is typically conducted in
two ways. First, previous research on change in function commonly represents
that change as a transition between two points in time. Hagestad (1990) challenges
that approach to representing change in functioning health as unrepresentative of
the overall experience an individual is likely to have. Indeed, with the availability
of panel data documenting the functioning status of older persons at multiple points
in time, there is no need to limit the window of observation of health change to
two points. Analyses of single transitions, while useful in their own right, suggest
that change in functioning is monotonic and unidirectional. Although it is difficult
to move much beyond a typology of change that corresponds to types of transitions
(that is, improving, declining, and no change), description of the details of
individual experience can enrich our understanding of what membership in these
simple categories might mean. In this study, functioning statuses over multiple
observations are linked to provide more information about individuals’ experiences
than is provided when only two observations are used.
Second, research on change in functioning typically does not make use of
information on prior functioning status. There is a prevailing analytic assumption
in the literature on changes in functioning health among older persons that an
individual’s history of functioning health prior to the first of the two time points
does not influence subsequent changes. In this research, previous change in
functioning is shown to improve our ability to predict certain changes in
functioning. These two elaborations, moving from transitions to patterns or
trajectories, and incorporating functioning history into a predictive model, allow a
number of salient research questions to be addressed. These are discussed in the
next section, and a discussion of the significance of these research questions
follows. This introduction concludes with a chapter-by-chapter overview of this
study.
A. Research Questions
A pair of complementary approaches is taken to represent change in
functional health over multiple observations. These new approaches to
representing change in functioning allow the following questions to be addressed:
(1) What patterns of change does a representative sample of older persons
experience over four observation points?
(2) How do proportions of older persons improving, declining, or maintaining
functioning differ when defined by the typical transition approach compared
with a "pattern" approach?
(3) Does knowing the something about the history of an individual’s functioning
change improve our knowledge or ability to predict their future functioning
change?
(a) Are predictors of change in functioning different if change is defined
as a trajectory over multiple observations instead of as change
between two times?
(b) Does prior change in functioning predict subsequent change in
functioning?
(4) Are the same characteristics associated with transitions to and from different
"levels" of disability (i.e. Activities of Daily Living and Instrumental Activities of
Daily Living)?
A broad array of characteristics hypothesized to be associated with change
in functioning are evaluated in the context of these new definitions of change.
Indicators of social, demographic, and economic characteristics, and a variety of
disease, condition, and other health indicators are included. An important
elaboration in this study is the distinction between related aspects of functioning
ability: impairment and disability. While disability in ADL is a widely used health
outcome and program eligibility measure, the notion of how disability fits into a
model of morbidity has not been formalized in a way that is universally accepted.
Building on models proposed by The World Health Organization (1984) and Nagi
(1991) which delineate pathology from impaired physical and cognitive
functioning, Verbrugge and Jette (1994) have proposed a model that further
distinguishes functional limitation or intrinsic disability (physical ability regardless
of situational requirements) from the endstate of actual disability (loss of
functioning that impedes the ability to perform a social role). These concepts are
discussed in greater detail in the subsequent chapter.
Pathology, impairment, intrinsic disability, and actual disability, while
related, differ in both their antecedents and consequences. Fried et al. (1994)
found that specific pathologies are variously associated with impairments. And,
social and environmental context may mediate the expression of impairment as
disability. The specification of the model includes a comprehensive set of
predictors and detail on diseases and impairments, and this allows the following
questions to be addressed:
(5) When a complete model including social, economic, demographic, disease,
condition and impairment information is used, what characteristics affect
functioning health?
(6) Does the conceptual distinction between impairment and disability hold
analytically?
(7) What is the relationship between diseases and disability?
B. Significance
Public health priorities for health related aging research have been focused
primarily on postponing mortality, and research has concentrated on life-
threatening diseases. More recently, concern for maintaining functional
independence throughout older age has prompted interest in chronic conditions and
the non-fatal outcomes of disease. Disability in old age is associated with
increased formal health care utilization, the need for personal assistance, loss of
autonomy and reduction in productivity.
This study builds on a well-developed literature which involves using
multivariate models to predict loss and recovery of physical abilities. The nature
of the elaboration is four-fold: (1) The representation of change in functioning is
elaborated to include information on status at more than two times; (2) several
related dimensions of health and health history are used to achieve a more
integrated analysis of the process of health change; (3) a model incorporating a
broader array of hypothesized predictor variables than are typically included allows
a better understanding of how each of these characteristics independently affects
the functioning outcome; (4) how these characteristics affect changes in functioning
are now better understood through comparisons of analyses using differing time
frames that provide a sense of the process by which older persons lose and recover
functioning.
These elaborations have significant implications for the research
community, practitioners, and the well being of older persons. The accuracy of
our understanding of the process by which functioning changes, and of the
characteristics associated with those changes, affects how we think about aging-
related health change. It also affects how loss of functioning is treated with
medical, rehabilitative or environmental interventions. A more precise
understanding of the process of health change is expected to lead to more effective
intervention. An important and often overlooked consequence of how we think
about changes in functioning with age is the effect of expectations of persons as
they age. While it is unlikely that the conceptualization of change in research in
any way directly affects general expectations, it is quite likely that it does
contribute to the paradigm under which practitioners operate. Certainly the
messages doctors and therapists give their patients influences the individuals’
perceptions of their chances of improvement or deterioration. Such expectations
may directly impact an individual’s will, motivation, and compliance with
prescribed or recommended measures.
C. Overview of the Chapters
In this section the issues and specific research questions dealt with in each
of the remaining chapters is described. This study is concerned with describing
and predicting change in physical functioning among older Americans over time.
Although a number of large, nationally representative data sets are collecting
information on ADLs, IADLs, and other measures of functioning ability among
panels of older persons, the study of change in functioning health is complicated by
a variety of analytic issues. These include the measurement, representation and
modeling of change in functioning in a panel sample as it moves through time,
losing people to nonresponse and death. While the goal of this study is to develop
a predictive model of change in functioning ability among older persons, these
analytic issues are an integral part of the research. Several chapters are devoted to
these topics.
1. Chapter II: The Conceptual Framework
There are two streams of research on change in functioning; one is
generated by sociologists, the other by epidemiologists and other medical
researchers. In this chapter, theory and findings from the two streams are
integrated to develop a conceptual framework for the present study. The resulting
conceptualization has two important features: First, the characteristics
hypothesized to predict change in functioning come from a synthesis of previous
research, and represent a more complete array of the potential determinants of
change than are typically included in studies of change in functioning. Second, a
shift in how change in functioning is conceptualized and represented is proposed.
This shift involves using information from individuals over multiple time points to
provide more detailed description and improved prediction of future change.
2. Chapter III: Data and Sample Dynamics
The data used in this study come from the Longitudinal Study on Aging, a
nationally representative sample of community-dwelling older persons who are 70
years of age and older in 1984. These persons are followed over six years, and
are reinterviewed at two-year intervals: 1986, 1988, and 1990. The
9
representativeness and observation plan is unique among panel studies of older
persons, and the approximately equally spaced interviews facilitate the description
of change at regular intervals. A detailed description of the design used to achieve
this sample and the characteristics of the resulting sample follow.
In addition to the information collected from the LSOA survey on health,
social, and demographic characteristics, linkages exist between the LSOA sample
and two other data sources, the National Death Index and Medicare Part A
(inpatient hospital) claims data. Details on these linkages and the use of these
auxiliary data are described in this chapter.
The Longitudinal Study on Aging is a panel study, and this means that the
same persons sampled for the initial interview are reinterviewed at the follow-up
waves. Because some of these people die, or survive but are not reinterviewed,
the characteristics of the sample may change over time. The final sections of this
chapter are devoted to describing the "sample dynamics" or changes in the sample
population because of death, loss to follow-up, or reinterview after loss to follow-
up.
3. Chapter IV: Evaluating the Effects of Nonresponse in the LSOA
Although panel data is increasingly available and holds great promise for
illuminating research on changes and the interrelationship between various
characteristics over time, analysis of panel data is complicated by the problem of
sample attrition. All studies using panel data must investigate the extent to which
loss to follow-up has occurred, and whether or not the loss is likely to bias analytic
10
results. This chapter presents an extensive exploration of the determinants and
consequences of loss to follow-up. While a number of characteristics are found to
be associated with loss to follow-up, there are minimal differences in data that
have been corrected to compensate for this differential loss and data that have not
been corrected. The lack of difference in results between "corrected" and
"uncorrected" data suggests that serious bias has probably not been introduced by
loss to follow-up among the survivors in this sample.
4. Chapter V: Using Items from ADL and IADL Scales
to Measure Change in Functioning
In the fifth chapter, issues regarding the measurement properties of items on
the Activities of Daily Living and Instrumental Activities of Daily Living scales as
measures of status and change are addressed. Specifically, the dimensionality,
scalability and the reliability of the dimensionality and scalability o f the items over
time are evaluated. Also, since several formats (variations in question wording) of
the ADL and IADL questions are commonly used, differences in the
dimensionality and scalability of the items by question format are examined, and
differences in rates of change indicated by the different question formats are
described. Previous research directs specific analytic inquiry into these questions,
and the results suggest which question format should be used and how responses to
the ADL and IADL items should be combined in the subsequent analysis.
11
5. Chapter VI: Patterns of Change in Functional Health
Because data now provide information on functional status over multiple
observation points, it is possible to look at sequences of status over time and
represent these as trajectories of change. Such an approach is taken in the sixth
chapter: Patterns o f Change in Functional Health. This approach provides a
number of useful insights. The increased richness o f detail on the actual
experiences of older persons over four observation points provides important new
descriptive information about the heterogeneity o f the experience of older persons.
Within this heterogeneity, individual variations, in both the degree o f change and
the direction of change an individual experiences, highlight the nonlinearity of the
sequences of status observed among this sample of older persons.
This approach also allows a comparison of change defined by transitions
across two points in time, and change defined by patterns across four points in
time. Thus, the following research questions are addressed:
(1) W hat patterns of change does a representative sample of older
persons experience over four observation points?
(2) How do proportions of older persons improving, declining, or
maintaining functioning differ when defined by the typical transition
approach compared with a "pattern" approach?
Results of the predictive analyses are also used in comparisons with the results in
the next chapter. There are several limitations to this approach, however. The
first is the cumbersomeness of the individual patterns of change and the loss of
12
information that occurs when the patterns are categorized to make them viable for
quantitative analysis. This approach requires that complete data exist for each
person in the sample at each of the multiple observation points, and the sample is
selected in such a way that those with worse health and functioning are not
included in the analyses. Consequently, the results are applicable only to the
sample persons who remain alive and respond at each of the interviews. And,
because statuses at the four interviews are linked to form the dependent variable,
change in predictive characteristics that vary over time cannot be meaningfully
incorporated into the analysis.
6. Chapter VII: Processes of Change in Functional Health
A different but complementary approach is taken to provide another view of
the process of change in functioning in this second analysis. A dynamic model,
which allows the incorporation of time-varying covariates and the inclusion of all
sample persons up to the time they become deceased1 , is used. While this
approach sacrifices in some ways the details of the individual history of change, it
provides an alternate means of including some of this information that may be
more flexible and more informative. This approach also capitalizes on the
1 This approach actually requires each observation to have three consecutive
interviews, and thus the restriction on the sample in this analysis is similar to that in
the analysis of patterns, since data at four interviews is available and all of the sample
members responded to the first interview. In a panel with more observation points,
a more inclusive subset of the original sample could be used than with the "patterns"
approach, since response at every interview is not required.
13
availability of data and provides a mechanism to test a key assumption about the
nature of changes in functioning ~ that change in functioning ability is dependent
only on one’s status at the beginning of the interval of observation. This
hypothesis corresponds to the question:
(3b) Does prior change in functioning predict subsequent change in
functioning?
This approach has the advantages of a traditional transition analysis, while also
incorporating information on the individual’s recent history of health change.
Results from this analysis suggest that knowing an individual’s previous
functioning usually improves the prediction of subsequent change.
This approach analyzes transitions between two points in time,
incorporating information from earlier interviews as predictive characteristics. In
order to provide a comparison with the analysis in the previous chapter, these
transitions are characterized generally as improvement, decline, and no change.
However, analysis of the individual transitions (between each level of functioning
status at one time and each level at the next time), provides more information
about how the characteristics included in the model differentially predict certain
transitions and not others. This part of the analysis responds to the question:
(4) Are the same characteristics associated with transitions to and
from different "levels" of disability (i.e. ADL and IADL)?
14
7. Chapter VIII: Discussion
The results of the two approaches to representing individuals’ health
experiences presented in Chapter VI and Chapter VII are compared and contrasted
in the final chapter. These comparisons address the question:
(3a) Are predictors of change in functioning different if change is
defined as a trajectory over multiple observations instead of as
change between two times?
Results from the multivariate analyses in Chapter VI and Chapter VII together can
be used to address the remaining questions:
(5) When a complete model including social, economic,
demographic, disease, condition and impairment information is
employed, what characteristics affect functioning health?
(6) Does the conceptual distinction between impairment and
disability hold analytically?
(7) What is the relationship between diseases and disability?
The two approaches characterize change in functioning over different lengths of
time, and this provides new insights into the relationship between the various
diseases and the process of functioning loss.
The final chapter concludes with discussion and summary of the results
focusing on the contributions of this research and the implications of this study for
practice and future inquiry.
15
II. CONCEPTUAL FRAMEWORK
A. Introduction
The purpose of this chapter is to establish a theoretical framework of
change in functioning among older persons. First, the relationship between various
individual characteristics and functioning change are discussed. W hile much effort
has gone into postponing mortality, until recently relatively little emphasis has been
placed on postponing or preventing disability (Kovar & Feinleib, 1991). With this
recent interest in intervention has come a plethora of studies seeking to determine
the causes of functional decline in old age. And, while most of these studies rely
on some theoretical justification for the specification of their models, there has
been a lack o f attention to the theoretical development of a model of change in
functional health.
This chapter reviews various ways of representing change in physical
functioning, and previous research on predictors of change in functioning is
discussed. Diverse bodies of research are integrated to develop a conceptual
framework for thinking about change in functioning. An important part of the
framework involves the conceptualization of the process of change in functioning.
This chapter has three parts that can be broadly identified as The Outcome,
The Predictors, and Introduction to Alternative Conceptualizations. The first
section begins with a discussion of the outcome of interest, physical functioning,
and the indicators used to measure physical functioning. Various approaches to
representing change in physical functioning using these measures of status are
given special attention. In the second section, previous findings regarding the
predictors o f change in functioning are reviewed and the linkages between these
predictors and functional health are discussed. In the third section of this chapter,
implicit assumptions about the nature of change in physical functioning are
questioned.
B. The Outcome: Physical Functioning
1. Conceptualizing Physical Functioning
Increasingly, researchers on health and aging are choosing measures of
physical functioning that capture levels of both health and autonomy among the
older population. The importance of fully understanding the dynamics behind the
prevalence of disability and dependency is highlighted by the fact that a number of
programs for older persons use these measures as eligibility or screening criteria.
How the notion of disablement (the process of becoming disabled) fits into a model
o f morbidity has not been formalized in a way that is universally accepted. The
medical model of disease has traditionally been concerned with the physiological
aspect of the disease process, and the outcome in the medical model is the
physiological manifestation of the disease. Recognizing this as a limited view of
the implication of morbidity for an individual, the World Health Organization
(1976) suggested a broader conceptualization of the implications of disease that
includes limitations in an individual’s ability for "role fulfillment" as a final
consequence of disease. Kovar and Lawton (1994) summarize this
conceptualization (see also Figure II-l):
1. An impairment is any loss or abnormality of psychological,
physiological, or anatomical structure or function.
2. A disability is any restriction or lack (resulting from an
impairment) of ability to perform an activity in the manner or within
the range considered normal for a human being.
3. A handicap is a disadvantage for a given individual, resulting
from an impairment or disability, that limits or prevents the
fulfillment o f a role that is normal (depending on age, sex, and
social or cultural factors) for that individual. (Kovar & Lawton,
1994, p. 58).
This conceptualization has not been widely employed in the literature, probably
because of the difficulty operationalizing the key constructs as described, and the
lack of clearly identified linkages in this conceptual model. However, a more
recent recommendation that "’loss of autonomy’ be used broadly as an endpoint in
epidemiological studies of the relationship among specific diseases, impairments
and disabilities’" (The World Health Organization, 1984 in Katz, 1987) reflects an
operationalization that has become widely used. Building on this model and a
conceptual scheme proposed by Nagi (1991), Verbrugge and Jette (1994)
recapitulate this framework, refining the interpretation of each step in what they
call the "disablement process." They define disability to indicate loss of
functioning ability that impedes the ability "to perform a specified social role"
(Haber, 1990 in Verbrugge & Jette, 1994). Functional limitation is a stage
between impairment and disability that refers to the individual’s ability "without
reference to situational requirements" (Haber, 1990 in Verbrugge et al., 1994).
Verbrugge and Jette operationalize these constructs as:
1. Pathology includes diagnosis of disease, injury,
congenital/developmental condition.
18
2. Impairments are dysfunctions in specific body systems:
musculoskeletal, cardiovascular, neurological, etc.
3. Functional Limitations are restrictions in basic physical and
mental actions, for example, the ability to ambulate, reach, stoop,
climb stairs, produce intelligible speech, and see standard print.
4. Disability refers to difficulty doing activities of daily life such as
personal care, household management, job, hobbies, errands, clubs,
and socializing with friends. (Adapted from Figure 2, Verbrugge &
Jette, 1994).
The precise relationship between "functional limitation" and "disability" has
been explored in considerable detail by Verbrugge and Jette (1994). They have
coined terms to differentiate absolute physical ability, or physical impairment, that
might be measured in a performance type measure (intrinsic ability), from physical
ability mediated by behavioral and environmental adaptations (actual disability).
This basic notion can be made more precise by including other factors mediating
the relationship between intrinsic and actual disability. Such factors, which may
either enhance or diminish an individual’s absolute functioning ability, may include
their expectations of loss of ability, their response to pain, fatigue, and the loss
itself, as well as the various modes of adaptation (Figure II-2).
Relationship of Disability to Other Dimensions of Health
The model of morbidity and its consequences presented above included
indicators of two health conditions that, in this conceptual framework, occur
conceptually prior to disability: disease, impairment, and limitations. While
disease, impairment, limitations, and disability are clearly related, and the various
measures are sometimes used interchangeably for these three concepts (e.g.
measures o f impairment or limitations are used to indicate disability and vice
19
versa), the substantive outcome of interest may require that a distinction be made.
For example, if a researcher is interested in whether or not an individual remains
able to perform personal care independently, measures o f disability and not
impairment or limitations should be used. A person who has limitations such as
being unable to climb stairs or being unable to see may or may not be unable to do
ADLs and IADLs independently. The following sections present a review of the
conceptual relationship between disease and disability, impairment and disability,
and limitations and disability.
Diseases. The association between diseases or pathological conditions and
impaired functioning status seems intuitively obvious: If nothing is wrong
physiologically, why would functioning be impaired? There is, however, an
imperfect relationship between functioning decrement and disease states. This is
because diseases do not each affect functioning in the same way. Some diseases
tend to cause death before ability is impaired, while others tend to manifest
themselves primarily in their impact on functioning. This imperfect relationship
also occurs because, within those diseases known to be associated with functioning
loss, the types of function lost vary considerably from disease to disease. Specific
diseases associated with functioning loss and the type of loss with which they are
associated are reviewed in the discussion of predictors of change in functioning
below.
The relationship between impairments, limitations, and disability. In both
of the simple conceptual models of disability described above, "impairments" result
20
from disease (pathology), and these lead to limitations that ultimately underlie any
disability. Just as the relationship between disease and disability is imperfect,
however, so too is the relationship between both impairment and disability and
limitation and disability. Measures of impairment may include abstract (i.e., not
situational) physical abilities such as strength, range of motion, dexterity, balance,
control, speed, and mobility, as well as physiological functions such as blood
pressure and peak oxygen consumption.
These indicators of physiological impairment generally have been found not
to provide appropriate markers for risk of disability as measured by ADLs and
IADLs. Gerety et al. (1993) find that among a very frail population, substandard
performance on such measures does not indicate underlying decrement in the
ability to execute ADL and IADL functions. However, as research accumulates,
specific impairment markers are being linked to ability to perform ADL and IADL
tasks. For example, Posner et al. (1995) find that peak oxygen consumption and
calf muscle strength are associated with ability in ADL among a sample of older
women. In that same study, a balance/gait test and the effect of strength in 8 other
muscle groups were found to be irrelevant.
Some batteries have developed performance tests of maneuvers to represent
components of physical abilities required to complete the various tasks on the ADL
scale. Among these, some scales have been designed specifically for the purpose
of indicating performance on ADL and IADL functions in a clinical setting. These
21
tests have been constructed to produce high levels o f correlation on a less frail
population (Siu et al., 1993).
There are two explanations for why measures of impairment and limitation
fail to correspond highly with disability as measured by ADL and IADL items.
First, there may be a tendency for persons to under-report disability. Second,
performance-based measures tend to collect information on "absolute" ability, or
ability of an individual to perform an isolated component of the physical maneuver
necessary to complete a task-item in the ADL series. W hile a person may be
unable to perform a physical test, in everyday life, alterations in how movements
are made and resources are used may allow the older person to continue to perform
the task approximately as they always have, and without the need for assistance.
Such adaptive behavior may be obvious, especially when the person has lost
functioning ability suddenly; or, it may be so subtle that the person is not aware
that he or she has made the adjustment.
The potential mediating effects of environment on the relationship between
intrinsic ability and actual disability have been described by Lawton (1982).
Lawton refers to the potential of environments to optimize a person’s abilities as
"person-environment fit." Environmental adaptations may be planned and obvious,
such as grab-bars in bathroom, or may be spontaneous and go unnoticed, such as
choice of a chair that is easier to get in or out of, placement of furniture to provide
support while crossing a room, or organization of spaces in the house to minimize
use of stairs or travel between rooms.
22
The mediating effects of physical environment and adaptive behavior are
examples of factors that make impairment and disability distinct constructs. This
assertion is supported by a recent study by Reuben, Valle, Hays and Siu (1995),
which finds a low and inconsistent relationship between self-reported, interviewer
assessed, and performance based measures. They conclude from these findings
that the different approaches to collecting the information result in different
constructs being measured. These results substantiate the distinction between
intrinsic and absolute ability proposed by Verbrugge and Jette (1994).
Operationalizing Constructs in the Disablement Process in This Study
In the present study, the terms "physical functioning," "physical ability,"
and "disability" all refer to ability in ADL and IADL tasks unless otherwise
specified. The Activity of Daily Living scale was developed by Katz and
colleagues (1963) for use among chronically ill older persons to measure recovery
of functioning. These measures typically include items indicating whether or not
an individual has difficulty bathing, dressing, eating, toileting, walking, and getting
in or out of bed or chair. These items are thought to represent the basic personal
tasks an individual needs to be able to perform in order to maintain independent
living at the most rudimentary level.
Instrumental Activities of Daily Living are generally thought to measure the
ability to live independently in the community. These measures indicate ability to
shop for one’s self, prepare meals, manage money, and use the telephone. This
scale was developed by Lawton and Brody (1969), and is expected to capture "an
23
individual’s functioning in the social world and the world outside the home"
(Kovar & Lawton, 1994).
In Verbrugge’s terminology ADL and IADL items indicate some functional
limitation as mediated through behavioral and, depending on question phrasing,
social and environmental adaptations. In other words, they measure a person’s
disability. Nagi type items, such as difficulty walking up a flight of steps,
reaching up over one’s head, and difficulty stooping, crouching or kneeling, are
measures o f functioning limitations.
2. Approaches to Representing Change in Physical Functioning
In studies of change in functional health between two periods, items from
ADL and IADL1 scales have been used both individually (Crimmins & Saito,
1993; Verbrugge, 1992), and in various combinations. Change has been measured
as movement between the state of reporting no difficulty with any of the activities
included in a composite measure to reporting difficulty with one or more of the
activities, or as movement from difficulty to no difficulty (e.g. Branch, Katz,
Kniepmann, & Papsidero, 1984; Palmore, Nowlind, & Wang, 1985; Rogers,
Rogers, & Belanger, 1992). Change is also measured as movement along a
summative scale indicating the number of activities (sometimes weighted) with
which the individual reports having difficulty or inability to perform (e.g. Jaegger,
1 In some of the research reviewed here (e.g. Crimmins, Hayward, and Saito,
1994; and Kaplan, Strawbridge, Camacho and Cohen, 1993) the Nagi-type items are
combined with ADL and IADL items to indicate disability.
24
Clarke, & Cook, 1989; Kaplan, Strawbridge, Camacho, & Cohen, 1993). These
approaches are sometimes combined in a hybrid approach with IADLs being
ranked less severe than ADLs and/or the number of activities with which an
individual has difficulty grouped into categories indicating ordered states of
functional impairment (Crimmins, Hayward, & Saito, 1994; Manton, 1988).
Rationales for choosing one of these approaches over another are evaluated in
Chapter V.
Dynamics of Change in Functioning
Most research on change in functioning ability focuses on transitions
between levels o f functioning status. Generally, these studies find that while the
preponderance of change occurs for the worse, a small proportion of the sample
experiences improved functioning. Crimmins and Saito (1993) used the baseline
and first wave o f follow-up interviews in the LSOA to consider the determinants of
functional improvement and deterioration in the ability to perform individual
Activities of Daily Living, Instrumental Activities of Daily Living, and Nagi-type
items among older persons. This analysis revealed that there is a considerable
amount o f gross change, both improvement and decline, in functional ability
among older non-institutionalized persons. Branch and Ku (1989) and Manton,
Corder, and Stallard (1993) examine transitions between levels of aggregate
measures of ADL and IADL functioning items, and Rogers et al. (1992) evaluate
transitions between difficulty and no difficulty. The proportions o f older persons
making improvements and declines in these studies vary because of differences in
level o f aggregation of the measure of functioning, and because of differences in
the interval length. Nevertheless, the sense is similar: Among older persons,
decline in functioning is neither universal nor precipitous; there is considerable
individual improvement in the general population, despite a separate, dominant
trend toward worse health.
Although the data used in each of these analyses often provides information
on functioning status at more than two points in time for each individual, in each
case the analysis focuses on transitions between only two points at a time. There
are, however, two important exceptions to this generalization. In a recent study by
Verbrugge, Reoma, and Gruber-Baldini (1994), diary entries provide continuous
data on changes in several dimensions of functioning over a one year period
following hospital admission for a small sample of persons 55 years of age and
older. The results of this study indicate considerable dynamism and heterogeneity
in patterns of change in functioning status. In another study, Maddox and Clark
(1992) map patterns of change in group means for a variety of subgroups of
interest (e.g. sex, years of education, income) over six interviews for the relatively
young (ages 58-63) sample from the Longitudinal Retirement History Survey. An
important finding from this study, echoing the theme of the general literature on
transitions, is the striking nonlinearity of change over time for the population
subgroups.
The existing analyses of transitions and patterns of change assume different
conceptualizations of the process of change. The analyses of transitions (as they
26
have been done) assume that, while functioning status at the beginning of an
interval will influence any subsequent change in functioning, all other things equal,
previous changes in health do not influence propensity for, or direction of, change
in functioning. This assumption implies that health change follows a simple
Markovian process. By contrast, the approach that attempts to follow individuals’
statuses over time implies that health history does matter — otherwise no new
information would be expected from linking successive observations.
C. Predictors of Change in Functioning
There are two streams of research on change in physical functioning, one
generated by sociologists, the other by epidemiologists and other medical
researchers. Sociologists have tended to emphasize social, economic and
demographic variables and general disease categories, while epidemiologists
emphasize the role of specific pathological processes on change in functioning,
while including demographic characteristics only as control variables. The
specificity of measures varies, with sociologists including a more comprehensive
set of indicators and more detail on socioeconomic characteristics, and medical
researchers including the additional detail on pathology and physiological
functioning. Yet, the general scope of hypothesized covariates is quite similar.
Among the range of characteristics included in this body of research, however,
there is considerable inconsistency in which characteristics are included in a given
analysis, with most studies neglecting one or more categories of indicators. The
two streams of research are similar, however, in the way change in functioning is
represented. Almost exclusively, change in functioning is represented as change
between two observations, or change between each o f several observations (always
two at a time). The few exceptions to this generalization have been discussed
above.
Correlates of functional change have been reported in a variety of studies.
This section discusses studies that use measures constructed from self-report
functional health items (ADL, IADL, and sometimes Nagi or Rosow-Breslau
items) as the dependent variable. There is, however, considerable variability in a)
the way change is operationalized (as discussed above), b) the length of interval
over which change is observed, and c) the inclusiveness of the set of independent
variables specified. These differences, and variations in sample characteristics,
may account for many of the differences in the findings.
Previous research suggesting the association of various social, economic,
demographic, and health indicators with functional health is reviewed below. In
addition to reviewing the findings, the section discusses why there are associations
between the characteristics and the outcome. As noted earlier, empirical studies of
functioning change often do not establish hypotheses reflecting the causal or
associative relationship between the predictors and the outcomes. Just as there is a
disciplinary slant on the scope and specificity of predictors included, the extent to
which relationships of-variables within models are discussed varies too. Although
the general model of predictors may be approximately the same from study to
28
study, the interest of the researcher directs which predictor characteristics are of
interest and deserve interpretation.
The goal of this section is to present an overview of characteristics that
previous research suggests are associated with change in functioning, and to
discuss why each o f these characteristics might have an effect. The reason for
doing this is twofold. First, it is important to have theory about how and why
each hypothesized predictor will effect the outcome in order to identify competing
hypotheses and facilitate interpretation of the results. Second, it is important to
establish a universe of characteristics that might effect the outcome of interest.
1. Demographic. Economic. Social, and Psychosocial Characteristics
Age
Age is always a significant predictor of decline in functioning, but it is not
always associated with improvement. Crimmins and Saito’s (1993) study of
change in individual functions find that decrement, but not improvement, in
functioning is predicted by more years of age. Age may reflect genetic
endowments (and aspects of genetic frailty may emerge over time), as well as the
accumulation of environmental assaults and previous episodes of poor health that
may have weakened the individual’s resilience. Put another way, advancing age
suggests the increased probability of impaired ability to maintain homeostasis.
Some of the effect of age may reflect conditions uniquely affecting the current
cohort of older persons.
29
The relationship between sex and functional health is complicated. While
prevalence o f disability is generally higher for women than for men, researchers
find little difference in incident disability, as indicated by aggregate measures of
ADL and IADL ability (Guralnik & Kaplan, 1989; Kaplan et a l., 1993; Manton,
1988; Strawbridge et al., 1992). When change in ability to perform the individual
tasks is evaluated, women appear more likely than men to decline in Nagi-type
ability and less likely to lose ADL and IADL ability (Crimmins & Saito, 1993).
Women and men do not differ in their chances of recovery of ability, however
(Crimmins & Saito, 1993).
Sex may carry effects of genetic differences, and also differential effects of
social roles, patterns of labor force participation, social relations, and differences
in health practices and health care utilization behavior. The precise extent and
nature of "true" sex differences and the cause of such differences is still uncertain.
Moreover, some of these possible sources of difference are characteristics that may
change with social culture and with cohort membership.
Race and Socioeconomic Status
Measures of socioeconomic status are variable in their effects. African-
Americans usually are more likely to make a transition to worse health (Angel,
Angel, & Himes, 199.1; Guralnik & Kaplan, 1989; Strawbridge et al., 1992);
Rogers, Rogers and Belanger (1992) do not find race to predict transitions in the
total sample, although race and sex appear to mediate the effects of other
covariates. Also, Crimmins and Saito (1993) find African-Americans more likely
to lose ability to perform some individual functions, but neither more nor less
likely to experience improvement.
Race differences in rates and causes of death are particularly pronounced
between whites and African-Americans, and there are important differences in the
relative prevalence of disease and conditions between African-Americans and
whites. Berkman, Singer and Manton (1989) for example, conduct a grade of
membership analysis for African-Americans and whites. This technique relies on
empirically constructed profiles that represent "ideal types". Each individual may
have partial membership in some or all of the "types." Berkman et al. find that,
although profiles for the two groups seemed to represent essentially the same levels
of disability, the disease processes underlying the levels of disability were
distinctly different for the two racial groups. While it has not been possible to
look for such differences between persons of Hispanic origin and others because
adequate samples of Hispanics have not been available, it is possible that such
differentiation exists with (and within) this group as well (Angel et al., 1991).
The source of observed differences remains largely unidentified, however.
Dowd and Bengtson2 (1978) suggest that racial variation in age-related decline is
2 Dowd and Bengtson found evidence for the double jeopardy hypothesis
(interaction of age and minority status) on some variables of interest including health.
Since then others, such as Ferraro (1987) have not found that the effects of race vary
by age. However, how and why race/ethnicity effects appear is not yet clear, and
further research is imperative.
31
modified by social environmental factors, and Markides (1989) suggests that
minority status is associated with the stress of lower socioeconomic status and that
these effects compound the negative effects of age. This "stress" has been
operationalized by House et al. (1994) as smoking, alcoholic drinking, weight,
social support, stress, and self-efficacy. That study suggests that differences in
health behaviors account for race differences in health and functioning.
Differences in the relationship between socioeconomic status and health by
age have been reported in House et al. (1994). The relationship between low
socioeconomic status and health is greater at younger ages, but diminishes over
time as prevalence of poor health among those of higher socioeconomic status
increases. Still, a number of studies on older persons still find differences in
health by various indicators of socioeconomic status.
While those with higher education are less likely to lose ability (Harris et
al., 1989; Rogers et al., 1992), Mor et al. (1989) have found education to be
important only for predicting changes in health for women. Being in or near
poverty at baseline (Branch & Ku, 1989) and having lower family income
(Guralnik & Kaplan, 1989; Kaplan et al., 1993) have also been found to predict
declines in functional health. Rogers et al. (1992) find poverty status associated
with declines only for white women, and with improvement only for white men.
While income is a commonly used measure of socioeconomic status, it is
often not a particularly reliable item on many surveys. Further, the frequent
finding of no association between income and health in old age may be because
32
income in old age does not represent lifetime conditions (and it may be these
longer term conditions and not short term conditions that impact health). Poverty
threshold, while a coarser measure in some respects, has the advantage of
capturing the "stress" factor hypothesized by Markides (1989) and House et al.
(1994) and is more likely to discriminate according to access to health care.
Insofar as those in poverty over 70 years of age are thought to have endured a
lifetime of poverty, this measure may be indicating early life conditions. For
women, however, loss of a spouse’s pension with widowhood often results in a
dramatic drop in income that may not be representative of earlier periods in life.
Wealth is probably a better measure than income because it will not be as closely
associated with retirement status as income. Problems may arise, however,
because of complexities and shifts in the causal relationship between health and
wealth with age. For retired persons, the use of previous occupational status poses
the same problems as income and may be precluded by lack of data. Therefore,
some measure of educational attainment is often included in models of mortality,
health and functional status and change as a "catch all" measure of socioeconomic
status, sometimes in place of an income measure, and sometimes in addition to it.
The association between education and health is most often interpreted as an
indication o f current socioeconomic status, often a proxy for healthy habits, health
beliefs, and health care utilization behavior. However, as educational attainment is
often determined by one’s early life chances, it may be that in analyses which thus
33
far have not been able to include direct measures of early life conditions, education
is capturing these effects.
2. Health Behaviors
Among the specific attributes o f socioeconomic status that some of the
above socioeconomic measures are thought to capture are health behaviors. When
data allow the inclusion of more direct measures of health behaviors, physical
activity (Branch, 1985; Mor et al., 1989; Simonsick et al., 1993), and having body
weight within normal range (Crimmins & Saito, 1993; Guralnik & Kaplan, 1989;
Harris et al., 1989), appear to be important predictors of loss versus maintenance
o f functioning. Health care utilization behaviors have been less frequently
included, but fit within the same theoretical framework.
Healthy behaviors are perhaps the source o f the greatest potential gain in
life expectancy and active life expectancy. Behaviors such as physical activity and
regular exercise, low fat diet, maintaining moderate weight, not smoking, and
excessive alcohol consumption seem to have relatively direct effects on either
maintaining or destroying the integrity of particular systems, in turn strengthening
or weakening resilience to specific diseases. Another important healthy behavior is
appropriate health care utilization. How persons use health care may be
determined in part by their perceptions of medical efficacy and their personal locus
of control; health care use is also constrained by their access to health care
(insurance coverage or adequate income/wealth to purchase services) and by
provider availability.
34
Psychosocial measures are increasingly being used with great reward in
research on the health of the elderly (Antonucci, 1990; Mor-Barak & Miller, 1991;
Mor-Barak, Miller, & Syme, 1991). Although the mechanisms of the various
influences are not clearly understood, it is increasingly clear that social support is
an important component in resisting the impact of adverse health events. For
example, whether social support exerts a direct effect, or works as a buffer, only
evidenced in times of crisis, is still a subject of much inquiry (Mor-Barak &
Miller, 1991; Mor-Barak, Miller, & Syme, 1991). Good family relationships do
not affect health, but bad relationships have a very adverse effect (Antonucci,
1990). Being married usually is found to decrease mortality rates (e.g. Lillard &
Brien, 1993).
3. Diseases and Conditions
A particularly salient issue in the study of functioning ability is the
relationship between changes in disability and morbidity. The complexities of this
relationship have been introduced above. One can infer that functioning abilities
are related to physiological status because of the association between level of
greatest functional impairment and mortality (Manton, 1988).
Studies that have considered the relationship of disease with functioning
ability fall into three categories. The first are studies largely from medical
sciences that are concerned with the etiology and consequences of a specific
disease on the outcome of functioning impairment. Such studies have focused on
vision and hearing impairment (Laforge, Spector, & Sternberg, 1992; Rubin,
35
Roche, Prasadarao, & Fried, 1994), the effects o f musculoskeletal disease
(Vanschaardenburg, Vandenbrande, Ligthart, Breedveld, & Hazes, 1994), and
more specifically arthritis (Verbrugge, 1992), stroke or cerebrovascular accidents
(Nakayama, Jorgensen, Raaschou, & Olsen, 1994; Smith & Clark, 1995), chronic
obstructive pulmonary disease (Carter, Coast, & Idell, 1992) and cancer (Kurtz,
Given, Kurtz, & Given, 1994).
The second type of study has the aim of predicting changes in functioning
and uses a general multivariate model that attempts to include all o f the
hypothesized causes o f functioning change. These studies typically include
indicators for basic sociodemographic characteristics, possibly some other variables
of interest, and several disease variables in a model predicting change in a
composite score of ADL and IADL function. Among these, vision impairment
(Mor et al., 1989), arthritis (Boult et al., 1994; Foley et al., 1990; Harris et al.,
1989; M or et al., 1989), diabetes (Mor et al., 1989; Pinsky et al., 1985; Roos et
al., 1991), stroke (Kaplan et al., 1993; M or et al., 1989), cardiovascular disease
(Harris et al., 1989); hypertension (Guralnik & Kaplan, 1989; Pinsky et al., 1985),
body/mass index (Guralnik & Kaplan, 1989; Harris et al., 1989; Pinsky et al.,
1985), heart attack (Kaplan et al., 1993), and cancer (Roos & Havens, 1991) have
each been found to have an adverse effect on physical functioning. Most of these
conditions.have also been found to be not significantly associated with functioning
decline in at least one study. That variation is most likely due to differences in
sample, study design, or outcome measure.
The third class of study is concerned with the association between diseases
and loss of specific ADL or IADL abilities. There are few studies that are
attempting to pinpoint this complex association. Crimmins and Saito (1993) use a
multivariate model including age, sex, race, and a variety of diseases and
conditions in parallel analyses predicting decline (change from having no difficulty
to having difficulty, and from having difficulty to having no difficulty) for each of
the ADL and IADL tasks. They find arthritis is associated with improvement in
using the telephone, but otherwise is not associated with ability in IADL tasks.
Arthritis does effect ability to eat, transfer from bed or chair, and walk. Having
had a stroke is likely to cause disability in shopping, managing money, and
walking. Heart disease is associated only with difficulty walking, while vision or
hearing loss is associated with shopping, managing money, using telephone, doing
light housework, eating, walking, getting outside, and getting to or using the toilet.
Being in the lowest quartile of the body-mass ratio distribution (underweight)
affects meal preparation and using the telephone. Being heavy is also associated
with difficulty preparing meals. This study suggests that different conditions are
associated with differential types o f changes in functioning.
In a related analysis, Fried et al. (1994) consider the effect of various
conditions on empirically defined groups of functions that seem to be lost together.
These groups o f functions are identified using factor analysis on ADL and IADL
items as well as others. One of these groups corresponds closely with the ADL
scale, including ability to use the toilet, dress, bathe and eat (but not transfer or
37
walk), and another with the IADL scale, including paying bills, preparing meals,
shopping, doing light housework, and using the telephone. Other groups of
functions reflected mobility and upper extremity ability. Among the conditions
considered, having had a myocardial infarction, stroke, arthritis, difficulty hearing,
and recent weight gain are associated with the IADL type group. Interestingly,
most conditions are associated with at least two of these task groups. Having had
a stroke, arthritis, or hearing problem are also associated with difficulty in the
ADL type group of tasks. A number of other indicators of physiological
impairment are also found to predict loss of functioning in this study.
Because research has only just begun to investigate the relationship between
disease and disability, very few details are known about this complex relationship.
And, although the specific way in which a particular disease will affect a
composite ADL or IADL score (or rather, the particular kind of functioning loss it
is likely to be associated with) is complex, each of the following diseases have
been shown in one type of study or another to have an effect on a composite ADL
or IADL score: sensory impairment (difficulty hearing or seeing), stroke or
cerebrovascular disease, heart disease, chronic obstructive pulmonary disease,
cancer, osteoporosis, diabetes, arthritis, and variations from normal weight.
D. Summary and Elaboration: Introduction to Alternative
Conceptualizations
Clearly, there has been considerable effort focused on determining the
causes of functional decline in old age. However, there has been a lack of
38
attention to explaining the linkages between the predictors and the outcome. In the
previous section, results o f studies that establish the empirical association between
functional health and a variety of indicators were reviewed. In this section,
characteristics previously found to be associated with change in functioning health
are summarized and the discussion now focuses on describing the specific causal
relationship between the predictive characteristics and functioning change. A
working model with operational definitions of these constructs is proposed for use
in the current study. This chapter concludes with an introduction to a
reconceptualization of how change in functioning is represented. This elaboration
involves shifting how we think about change from transitions or movement between
two observations to sequences of change over multiple observations. It is argued
that a more realistic way to think about how and why a person’s functioning
changes involves knowing something about their past functioning. Such
information is typically not included in studies of functioning change.
1. Summary of Hypothesized Predictors
Socioeconomic and Demographic Characteristics
A schematic diagram synthesizing and elaborating the research reviewed in
the previous section is presented in Figure II-3. Certain attributes have been
shown to be associated with increased propensity to physical frailty in old age.
Age is expected to capture accumulation of experiences and a gradual loss of
homeostasis. Sex differences in health may arise because of physiological
differences, differences in environmental exposure (e.g. differences in exposure to
39
occupational hazards), differences in health behaviors, especially health care
utilization, systematic differences in allocation of treatment to men and women,
and possibly differences in how physical ability is reported.
Race is operationalized as African American or not African American; the
sample size for other racial groups is too small for reliable analysis. To attempt to
tease out the extent to which effects by sex and race are due to differences in
socioeconomic status (SES) and health behaviors, education and poverty status will
be used as SES indicators. Wealth measures are unavailable in this data, and
family income is unreliable and a poor indicator of economic well-being among
this largely retired sample.
Exactly what educational attainment is expected to indicate is often
ambiguous in studies of health and aging. It is typically included as an indicator of
socioeconomic status and may be capturing effects of early childhood conditions,
occupational history, health behaviors, or access to health care (by way of
income). In this study, several indicators of health behaviors are included:
exercise (whether or not an individual walks a mile at least once a week), and
weight (the conventional body-mass ratio is used to indicate whether or not an
individual’s weight falls within the "normal" range).
Although individual monetary resources are hypothesized to affect health
behavior, and, in many cases, access to health care, using indicators of income is
complicated. An individual’s level of income may not be the same in retirement as
it was in the years prior to retirement, when retirement health may be partly
40
determined. Here, an indicator for poverty status is included. Persons who are
below the poverty threshold in retirement were probably not very well off prior to
retirement. This generalization does not necessarily hold for women whose
financial status in old age may depend on the survival of the husband (and on
receipt of his pension and social security benefits) more than on his retirement
status.
Disease. Conditions, and Impairments
Two kinds of disease and condition indicators are included. These indicate
history of disease and conditions, and also serious episodes that occur during the
course of the six-year survey. This is discussed in more detail in subsequent
chapters. The diseases and conditions included are those suggested by previous
research and those for which there are enough observations to permit analysis.
These are: difficulty with hearing; difficulty with vision; diabetes; arthritis;
osteoporosis; heart disease; cerebrovascular disease and stroke; cancer; and chronic
obstructive pulmonary disease. Although introduced as a measure of health
behavior, body-mass ratio may also be considered an indicator of a pathological
condition. For example, cancer in its final stages is often accompanied by severe
wasting (cachexia).
Functional Limitations
Drawing on Verbrugge and Jette’s (1994) conceptual association between
functional limitation and disability, several Nagi-type items (walking a quarter of a
mile; walking up 10 steps without rest; standing or being on feet for about two
41
hours; sitting for about two hours; lifting or carrying 10 pounds; stooping,
crouching or kneeling) are also included as predictors o f change in functioning.
These items represent a state of physical decrement conceptually prior to ADL and
IADL disability.
2. Reconceptualizing the Representation of Change in Functioning: An Introduction
While considerable attention has been given to the predictors of change in
functioning health in the literature, very little emphasis has been given to the
conceptualization of the process of change in functioning. The conceptualization of
change in functioning is characterized by two features in most analytic studies.
These two features are the representation of change in functioning as transitions
between two points in time, and the assumption that functioning status at the
beginning of an interval sufficiently captures whatever needs to be known about the
prior history of functioning in the individual to predict change in that interval of
observation. This raises the question of whether or not knowing a person’s history
of changes in functioning improves the prediction of future changes. Specifically:
(1) does the change (or lack of change) a person experiences over several
observation points look different from change over two points in time? (2) Are the
predictors of change over several observation points different from those of change
over a single interval?
It has been noted that most previous research on change in functioning has
employed analytic strategies that restrict the evaluation of change to transitions
between two points in time. Because of the conceptualization of the maintenance
42
of functioning ability as an ongoing dynamic process, it is proposed that
trajectories over multiple time periods are required to understand the functional
health o f an individual. Hagestad (1990), reflecting on the recent surge o f interest
in models that use transitions as the basis for analysis, suggests that more attention
be given to describing trajectories of individual experience over time. While
transition analyses tell us something about the probabilities of change in status,
they do not tell us the history of health changes an older person is likely to
experience over time, and as such, they neglect the variability of individual
experience. In the sixth chapter, individual histories of functioning status are
constructed over four survey interviews spanning six years.
The second feature common to studies of change in functioning is that they
control for initial functioning status. Analytic models of change in functioning
typically include functioning status at the beginning of the interval as a "control"
for level o f ability. The implicit hypothesis is that persons who are either more or
less able will be more or less likely to make transitions of a particular type.
Including information on only status at the beginning of the interval of analysis,
and not before this, assumes that initial status captures the relevant information
about an individual’s health for predicting likelihood of next change in health.
This assumption may be untenable. In the seventh chapter, a transition approach is
adopted so that this assumption can be evaluated. This leads to an analytic
approach that has the advantages of a transition analysis and also incorporates
information from individual histories.
This study examines in greater detail several features of the model
described above. Key features of that model are (a) the inclusion of a broad set of
characteristics that have been shown to predict change in functioning over an
interval, (b) the inclusion of precursors to physical disability, including diseases,
conditions, and impairments, (c) and also the inclusion of previous health status to
predict subsequent change. O f utmost importance to this endeavor is the
incorporation of information from multiple observations of health status. Two
analytic approaches are taken which do this in different ways. The two approaches
yield unique and interesting results. In combination, however, they provide an
added dimension to our understanding of how the hypothesized characteristics,
particularly those indicating disease states and conditions, affect functioning health.
Since one approach defines change as a six-year trajectory and the other uses a
two-year transition, differences in the effects of the predictors create a sense of
timing and of the period of risk introduced by these factors. This dimension has
not been visible with previous approaches.
The following two chapters describe the sample and evaluate the panel for
significant changes introduced by loss to follow-up. The fifth chapter examines the
measurement properties of the Activity of Daily Living and Instrumental Activity
of Daily Living items that are used to form the indicators o f disability status in this
study. The sixth and seventh chapters present findings from two approaches to
representing change in functioning over time, the first using a six-year period of
change, the second using transitions over an interval. In the eighth and final
44
chapter, the results from these two approaches are summarized, and the
implications and significance of the findings are discussed.
Figure 1 1 -1 : World Health Organization Model of Disease C onsequences
Disease ♦ Impairment ■ > Disability — - > Handicap
(exteriorized) j (objectified) | (socialized)
Kovar and Lawton (1994)
ON
Figure 11-2: V erb ru aae's C onceptualization of Intrinsic an d Extrinsic Ability. E m bellished
Behavioral R esponse to:
Intrinsic Ability
p ain
fatigue
lo ss of ability
Social Conditioning
expectation of loss of functioning
gender role expectations
Adaptations
environmental modification
special equipm ent
personal assistance
behavioral modification
Absolute Ability
Figure 11-3: Synthesis and Elaboration of Theoretical Models Implicit in the Literature
Genetics
propensity to frailty
disease markers
Morbidity
frailty
diseases
conditions
Functional
Limitations
Absolute
Ability
Social Environment
sex
race
Education
Health Behaviors
healthy behaviors
health care behaviors
Wealth/Poverty
Work Environment/Occupational Hazard
oo
III. DATA AND SAMPLE DYNAMICS
A. Introduction
In this chapter, the data used in the analyses presented throughout this study
are described, and an evaluation of the dynamics of sample attrition is presented.
Because the research questions are specifically concerned with change over time, it
is necessary to use longitudinal data from a panel sample so that characteristics of
individuals may be monitored as they age. While panel data is essential for such
studies, it also presents several special analytic issues. Losing panel members to
follow-up over the several observation points may change the original
representative distribution of the sample. Other issues arise as a result of
particular choices o f analytic strategies; these will be discussed as they become
relevant during those analyses.
B. Data
The Longitudinal Study of Aging (LSOA) is the primary dataset utilized in
this analysis. The survey begins in 1984 and three follow-up interviews are
conducted at approximately two-year intervals spanning six-years: 1986, 1988,
1990.
The sample for the LSOA is based on the Supplement on Aging to the
National Health Interview Survey. The NHIS is a multistage complex household
survey conducted throughout the year. Sample members are interviewed in their
homes by U.S. Bureau of the Census interviewers. All persons in NHIS-selected
households over 55 years of age were eligible for the Supplement on Aging in
49
19841 . The Longitudinal Study on Aging is a sample o f persons who were 70
years of age and older and had been included in the 1984 SOA. In 1986, cost
constraints required that the LSOA select a subsample of those eligible people.
This sample is based on three stages of selection: (1) All households with persons
80 years o f age and older were retained and the individuals 80 years o f age and
older and any relatives who were 70 years or older were interviewed; (2) Among
the households with persons aged 70-79, all African-Americans and persons of
Hispanic origin were selected; (3) Finally, half of the remaining households (white,
non-Hispanic, aged 70-79) were retained and all persons over 70 years in each of
those households were included. This selection process resulted in a final sample
of 5,1512.
When a sample member is unable to respond because of a physical or
mental health problem, proxy responses are accepted. Additional detail of this data
set and the survey procedures are provide in two NCHS publications (Fitti &
Kovar, 1987; Kovar, Fitti, & Chyba, 1992).
1 O f this eligible sample, only half of those aged 55-64 were selected for the SOA,
but all of the persons over 64 years were retained.
2 Among the 7,541 persons interviewed in the 1984 SOA, those still living were
eligible for a reinterview in 1988 and 1990. Some of these people were not
interviewed in 1986 and the resulting unequal intervals for those not selected in 1986
seriously complicates longitudinal analysis. Therefore, only the 5,151 sample eligible
for interview at each wave are used in this analysis.
50
1. Longitudinal Design
As the sample is not replenished at each wave, by 1990 the sample is a
group 76 years of age and older. All four interview waves will be used in the
analyses. The maximum potential sample consists of 5,151 people for whom
information is potentially available at four dates, with change over three intervals.
Because people die and are lost to follow-up, the number of people from whom
data are collected at each wave is less than the potential sample size.
The LSOA has a number of features that make it ideal for use in this study.
This is the only four-wave longitudinal survey of a nationally representative sample
of the community-dwelling older population. In addition, these four waves of data
were collected from approximately equally spaced intervals. These two features
allow the analysis of transitions over multiple intervals of equal length. The other
multi-wave survey of a nationally representative older population is the National
Long-Term Care Survey (NLTCS). That survey now has existing data for three
waves and the promise of a fourth. However, in contrast to the LSOA, all survey
intervals are of different length. This complicates analyses of patterns and
transitions. In addition, the NLTCS does not give a detailed survey to community
members who, at the initial screening in 1982, are reported to be without a
disability. This means that many fewer details on the health status and changes in
status of community dwelling nonrdisabled population can be gathered from this
survey.
51
2. Linked Data
The LSOA is the primary data source for this study, but its utility is
enhanced by linkages with two other data sources, the National Death Index and
the Medicare Part A Claims Data.
National Death Index
O f importance to this study is the fact that deaths for LSOA sample
members are tracked with the National Death Index as well as by using contact
persons whose names were solicited at the first interview (Fitti & Kovar, 1987).
National Death Index (NDI) data are collected for sample members and appended
to survey files. This results in nearly complete knowledge of the vital status of
panel members. A sample person is assumed to be dead if a contact person has
reported that person as deceased or if that person has a "good match" with the
National Death Index (presumed deceased) and did not respond to any surveys
subsequent to the date of death. The process of matching the NDI records with the
LSOA files been completed using the most recent update of the Index.
Medicare Part A Claims Data
In addition to linking LSOA members to the information on their death
certificates, members’ LSOA records are linked to their Medicare A records for
the duration of the survey (October 1984 through August 1990). The Medicare
Part A Hospital Record files include length of stay, date of discharge, diagnostic
codes, and cost of stay. Diagnostic codes and cost of hospital care are included in
later analyses.
52
3. Distribution of Demographic Characteristics
The distribution o f this sample by gender reflects the national proportions
for this age group. Approximately 60% of the sample is female. An adequate
representation of both genders is critical in research on aging, and especially in the
area o f functioning health. Prevalence of diseases and conditions, mortality rates,
characteristics of social support networks, health care utilization behaviors, and
types of functioning impairments are some o f the characteristics of the older
population that differ by gender. Why these differences exist is still largely
unexplored.
The LSOA prioritized the inclusion of minority sample members. The
initial sample consisted of all individuals who were 55 years of age and older
living in a household where an NHIS survey was completed. In 1986, when
budgetary limitations required reducing the sample, all African American or
Hispanic persons and their relatives over 70 years of age were retained in the
sample. Nevertheless, only 3% of the sample members claim Hispanic origin and
9% are African American. Race will be included in all multivariate analyses.
However, due to the limited representation of other minorities, only the effects of
being African American versus being white or other minority can be compared.
It should also be noted that the LSOA prioritized including the oldest-old in
its sample. The sampling strategy used to achieve a large sample of persons over
age 80 was described above. Too much research on the elderly still
inappropriately groups older ages, includes primarily the young-old, or has a poor
53
representation of the very old. In this study, age is included as a continuous
variable.
C. Sample Dynamics
1. Introduction
Although the increasing availability of panel data has been a great boon to
researchers studying characteristics that change with age and over time, analysis of
panel data is complicated by the problem of sample attrition. All studies using
panel data must investigate the extent to which loss to follow-up has occurred and
whether or not the loss is likely to bias analytic results. Surveys that collect panel
data over time are especially vulnerable to such distortion, as accumulated effects
are likely to introduce bias of greater magnitude. Likewise, if rates of attrition or
characteristics of attriters vary over time, there will be implications (and
complications) for both the quality and magnitude of bias.
This chapter examines the level of nonresponse in a four-wave sample that
is initially representative of the noninstitutionalized U.S. population 70 years of
age and over. A description of how loss to nonresponse and death contribute to
the decrement of the sample over time is presented. In the following chapter, the
effect of loss to follow-up on the representativeness of the sample is evaluated.
2. Sample Dynamics in the LSOA
All studies using panel data must investigate the extent to which loss to
follow-up has occurred and whether or not the loss is likely to bias analytic results.
This section begins with a description of how loss to nonresponse and death
54
contribute to the decrement of the sample over time in the LSOA. First, aggregate
outcomes are presented for the entire sample for the four waves. The same
information is then presented in more detail in the form o f transitions among
response categories and death. This gives a better idea of the dynamics producing
the aggregate outcomes. Individual patterns of response over the four waves are
then examined.
Description of Sample Dynamics
The number of respondents diminished at each successive wave (Table III-
1A). O f the original 5,151 respondents, 80 percent were interviewed in 1986, 64
percent in 1988, and just over half of the sample was left at the last wave in 1990.
Two-thirds of this loss is due to death. By the second wave, 13 percent of the
original sample had died; by the fourth wave more than one-third was dead. Even
in this older sample, the percentage who died between interviews among those
known to be alive at the preceding interview was fairly constant across the three
intervals (Table III-1B). Between the first and second interviews, twelve percent
o f the sample died; between the third and fourth interviews this was about 15
percent.
The percent of the original sample not responding to the survey increased
with each wave. Among those not known to be dead at each wave (Table III-1C)
the likelihood of nonresponse increased from almost 9 percent at the second wave
to 19 percent at the fourth wave. The percentage becoming new nonrespondents
(Table III-1D) does not increase nearly as dramatically, indicating that the
55
likelihood of not responding to future waves is much higher for those who are
earlier nonrespondents.
W hile Table III-l indicates the overall level of nonresponse and death in the
sample, the likelihood of changing or retaining response status is better shown in
Figure III-l, which indicates the outcomes on successive waves for those who are
interviewed and those who are nonrespondents at the earlier wave. At each wave,
75 to 80 percent of those interviewed at the previous wave were interviewed at the
next wave. At each successive wave, about one-third of those who were
nonrespondents on the previous wave are interviewed, while about 50 percent
remain nonrespondents. While these percentages are fairly constant over the
waves, the numbers to which they are applied are changing and this results in the
build-up of nonrespondents and the decrease in respondents at later waves.
Table III-2 shows the distribution of selected baseline characteristics for the
original representative sample and that sample after it has been selected on
response over the six-year longitudinal survey period. This simple presentation
suggests that loss to nonresponse and death make the sample younger, more likely
to be female, more likely to be a home-owner, and healthier. And, although the
selected sample includes more persons in households that have multiple sample
members, this selection does not affect the sample composition by race, living
arrangement or level of educational attainment.
56
Patterns of Nonresponse
There are 15 possible patterns of non-response (Table III-3). Because there
are three potential waves of response (3 follow-up interviews) in this analysis,
there are four general categories of response patterns to consider: response at all
interviews (IIII3 ), response and then attrited at some subsequent interview (INNN),
present at the initial survey, lost, and then found (INII), or possibly lost again
(NINN).
Consider the last two patterns, INII and ININ. The final pattern may
simply be an elaboration on the lost and then found pattern, or it could be a
combination of that pattern and an attrition pattern. Likewise, the pattern that is,
for the purpose of the survey, considered attrition (INNN), may simply be a right-
censored lost and found pattern — that is to say, if the survey continued, this
individual might eventually transition to respondent status.
The pattern of response of individual cases presented in Table III-3 gives a
fuller picture of how sample nonresponse affects our ability to trace changes in
individual’s characteristics across interviews. Eighty percent of the original
respondents are interviewed at every wave before death. Another 10 percent of the
sample is only missing at one interview and assumed to be still alive. Only four
percent of the sample are never interviewed again after the initial interview, and
one-third of these are dead before the fourth wave. Thus, even with this level of
3 1 = interviewed; N = not interviewed.
57
attrition, the changes in individual characteristics can be traced for most living
members of the sample.
D. Discussion
As noted above, nonresponse is an issue in all panel surveys. In the first
household interview of the National Medical Care Utilization and Expenditure
Survey (NMCUES 1980), 10% did not respond. By the fifth interview, conducted
15 months later, 22.1% of the sample were nonrespondents (Wright, 1983). About
5 % of the Survey of Income and Program Participation sample did not respond to
the first interview. By the ninth wave, 36 months later, this rate had accumulated
to 22.3% (Kasprzyk & McMillen, 1987). Nonresponse to the annual interview of
the PSID was high in the first wave (24%), then leveled off to about 2% at each
interview, accumulating at about 44% in the twentieth wave (Hill, 1992).
While there are a number of reasons the loss to follow-up among older
persons might be expected to be higher than that in the rest of the adult population,
loss to follow-up in surveys that are of only older persons compares quite well
with these general population surveys. Corder and Manton (1991) compare rates
of nonresponse in four longitudinal studies of older persons and report low
nonresponse rates of about 4% among those who had previously completed a
screening interview in the National Long Term Care Survey, and a high of 25% in
the NHANES I follow-up which required a physical exam. Both the LSOA and
the National Nursing Home Survey fall within the moderate range. The LSOA
compares favorably with other surveys of older persons and with general
58
population surveys with an initial nonresponse rate of about 9% , accumulating over
the four interviews (six years) to about 19%. The slightly younger male sample of
the Retirement History Survey, aged between 58 and 63 at baseline in 1969, was
diminished 11% by the second interview two years later by death,
institutionalization and nonresponse. Forty-six percent had been lost by the
seventh interview wave (Burtless, 1987).
Examination of the process of decrement to the sample in the Longitudinal
Study on Aging reveal considerable stability in the rates of loss to both death and
nonresponse. The assessment of sample dynamics suggests that a critical
component in the ultimate level of nonresponse in a panel is the accumulation of
nonresponders over time. And, although patterns of nonresponse and subsequent
response are the least common pattern, such sequences make up a non-negligible
six percent of the patterns. Thus, both preventive measures to keep sample
persons from being lost to follow-up and efforts to reinterview nonrespondents
should be pursued. Some nonresponse is inevitable, and efforts to implement
rigorous field protocol to ascertain the precise reason or circumstance for the
nonresponse should be made.
In addition to the decrement in sample size, the relationship between the
cause of the loss to follow-up and the characteristics of interest to the researcher
present special problems. When loss to follow-up is not random and is related to
outcomes of interest, bias is introduced. The selection of the sample may result in
confounded associations (Schaie, Labouvie, & Barrett, 1973; Siegler & Botwinick,
59
1979) and distorted prevalence estimates and transition rates (Corder, Woodbury,
& Man ton, 1992). For example, if an analysis of functional health change among
the elderly is undertaken using a sample in which there is significant attrition that
is related to functional health, the sample is selected with respect to the
characteristic of interest. Analysis o f functional health in that case will result in
biased estimates of the correlates of health change and the level of health o f the
sample. In the subsequent chapter, characteristics of nonresponse in the LSOA are
identified and the potential bias introduced by nonrandom nonresponse in this
survey is assessed.
60
Figure III-l: Transitions in R espondent Statu s Over 4 Waves in the LSOA
Wave 1 Wave 2
Wave 3
Wave 4
1984
1986
1988
1990
Interviewed
3 1 7 5 (7 7 , 1 8 % )
Interviewed
2 5 0 1 (7 5 . 4 7 % )
4114
3314
(79.9%) (73.5%)
Interviewed Nonresponse
2 0 3 (5 1 . 7 0 % )
/
Interviewed
Nonresponse
(12.1% )
2 6 9 (4 9 . 3 6 % )
Nonresponse
618
(16.0%)
V, \l
Vi \ t
\ ' d \ \ e
v
Dead
566
(14.7%)
Table III-l: Descriptive Statistics for Nonresponse in 4 Waves of the LSOA
1 1 1 -1 A. Frequency and Percentages of Persons Not Interviewed at Each Wave by Death and Nonresponse
Wave 1 Wave 2 Wave 3 Wave 4
1984 1986 1988 1990
n % n % n % n %
Interviewed 5,151 o
o
o
• t*
79.9 3,314 64.3 2,676 52.0
Lost to Death
. . . .
----- 645 12.5 1,291 25.0 1,857 36.1
Nonresponse
. . . .
----- 392 7.7 546 10.5 618 12.0
Total 5,151 100.0 5,151 100.0 5,151 100.0 5,151 100.0
111-1B. Percent of Those Alive on Previous Wave Not Interviewed on Next Wave Because of Death and Nonresponse
Wave 2 Wave 3 Wave 4
Lost to Death 12.5 14.3 14.7
Nonresponse 7.6 12.1 16.0
111-1C. Percent of Those Alive Who are Nonrespondents
Wave 2 Wave 3 Wave 4
8.7 14.1 18.8
III-l D. Percent of Those Alive Who Become Nonrespondents for the First Time Among Those Who Were Not Previously Nonrespondents
Wave 2 Wave 3 Wave 4
8.7 9.7 11.1
C T \
to
Table III-2: Comparison of Characteristics at Baseline for the Total
Sample and for the Sample Selected to Include only Those Who Respond
at Every Wave
Female
African American
Lives alone
Owns home
Health status
Poor
Excellent
Number of functioning
difficulties
None
One
Six
Thirteen
Other sample person in
household
Mean age
Mean years of
education
Mean number of
functioning
difficulties
Full Sample
n=5,151
64%
11%
37%
76%
13%
15%
60%
13%
2%
1%
36%
78
10
2
Selected Sample
n=2,431
67%
11%
37%
80%
7%
19%
71%
1 2%
1 %
0%
37%
77
10
1
63
Tablj^i i II^3je_Resgonse=Patterns_for_Three= J7ave|_of_Follow^u£
I=interviewed N=nonresponse D=dead
pattern frequency percent
III 2431 47.2% 47.2%
IID 442 8.6
ID. 597 11.6 32.7%
D.. 645 12.5
U N 302 5.9
INI 134 2.6 9.9%
Nil 70 1.4
INN 128 2.5
NIN 47 0.9 4.2%
NNI 41 0.8
IND 80 1.6 2.0%
NID 22 0.4
ND. 50 1.0 1.4%
NND 21 0.4
NNN 141 2.7 2.7%
100.1 100.1
Completed all interviews
Interviewed until death
Two more interviews
still alive
One more interview
still alive
One more interview
dead
Never interviewed again
dead
Never interviewed again
presumed alive
64
IV. EVALUATION OF THE EFFECTS OF NONRESPONSE
OVER FOUR WAVES OF THE LSOA
A. Introduction
In the previous chapter, the loss of sample members to nonresponse and
death was described. W hile any decrement in the sample reduces size and may
thereby compromise the adequacy of the sample, special problems arise when the
loss is not random. If certain characteristics make a person more likely to become
a nonrespondent, and if these characteristics are associated with the outcome of
interest in a particular analysis, then biased estimates may result.
Samples of older persons may be more vulnerable to selection effects from
nonresponse than younger samples because the proportion lost to follow-up tends to
increase with age (Rodgers & Herzog, 1992), and because the differences between
participants and nonrespondents may be greater among older persons than younger
(Schaie et al., 1973). Attrition occurs because respondents move or die and cannot
be located or because they refuse to participate. Loss may be more common
among the old because they die more frequently, have worse health, and have
more leisure time than those of other ages. Older respondents who are no longer
at the same address could have entered a nursing home or other institution, leaving
no household members behind to report their whereabouts. Those who refuse to
respond may be physically frail, cognitively impaired, or busy with social
activities.
65
In this chapter, a predictive analysis is used to determine which
characteristics are associated with nonresponse in this sample o f older persons.
Then, these characteristics are used to construct a weight that compensates for the
bias introduced by nonresponse. Comparing data that have been corrected for
nonresponse with data that are not corrected indicates how much bias has been
introduced by the nonresponse. In the next section, a review o f literature on
characteristics predicting nonresponse is presented to guide the analysis. Analytic
sections follow.
B. Background
Wave nonresponse is a special case of unit nonresponse. Unit nonresponse
occurs when a sample element (e.g. a person or household) does not respond at an
interview; wave nonresponse occurs in the case when no data are collected for a
sample element for any given wave (interview) in a panel survey, on the condition
that at least one wave of data is obtained for that sample element (Lepkowski,
1989). The term "attrition" refers to the case when a sample person is lost from
the panel for any reason and never returns. Attrition nonresponse is thus
differentiated from other kinds of nonresponse because the sample person either
cannot return (for example, they have died or left the eligibility set for the
sample), or because survey protocol does not include reinterviewing persons who
have ever been nonrespondents. This evaluation is concerned with the effects of
wave nonresponse.
66
1. Individual Characteristics Affecting Nonresponse
There are several factors that are consistently implicated as causes or
covariates of loss to follow-up among older samples. Many studies have reported
that nonrespondents are likely to be in poorer physical health (Chyba & Fitti,
1991; Manton, 1988; Rodgers & Herzog, 1992; Speare, Avery, & Lawton, 1991;
Streib, 1982) or have lower levels of cognitive functioning (Kalton, Lepkowski,
Montanari, & Maligalig, 1990; Schaie et al., 1973; Siegler & Botwinick, 1979).
In the Longitudinal Study of Aging, those more likely to become nonrespondents
have been shown to have a greater number of functional difficulties and worse self-
reported health (Chyba & Fitti, 1991; Speare et al., 1991). In the National Long
Term Care Survey, nonrespondents to the initial survey subsequently had higher
rates of mortality and institutionalization than continuing respondents (Manton,
1988).
In general, those who are of lower socioeconomic status are also more
likely to be nonrespondents. Those with fewer years of education (Streib, 1982),
lower income (Chyba & Fitti, 1991; Streib, 1982), and those who do not own their
homes (Kalton, 1990; Speare et al., 1991) are more likely to become
nonrespondents.
There are several other factors that have somewhat more ambiguous effects
on propensity to respond. For example, household composition and family
composition have been neither consistent in their effects nor in their level of
significance. Not being married is sometimes associated with nonresponse (Speare
67
et al., 1991) and sometimes not (Streib, 1982). Likewise, the response rate among
African Americans is sometimes lower than among whites (Foley et al., 1990), but
sometimes not (Streib, 1982). Men are sometimes found more likely to be
nonrespondents (Jay, Liang, Lui, & Sugisawa, 1993; Kalton et al., 1990; Streib,
1982), but sex has also been found to be insignificant (Adams et al., 1990).
Although age is almost always a predictor of the likelihood of nonresponse (Adams
et al., 1990; Rodgers & Herzog, 1992), its effect is sometimes nonlinear (Jay et
al., 1993). The inconsistency in the effect of these variables likely results from
differences in model specification or differences in the composition of the sample
population.
Change in Characteristics of Nonrespondents over Time
When several waves of data are available, it becomes possible to consider
how the characteristics of attriters vary over time. In a study o f nonresponse in a
sample of individuals 65 years of age and older, Norris (1985) finds that over five
waves, those who become lost to follow-up in earlier waves are in better physical
and psychological health than those who become non-respondents at later waves.
And, although type of attrition is associated with the time at which the respondent
first becomes lost to follow-up, an analysis of type of attrition by time of attrition
indicates that there is no effect of time independent of attrition type.1
1 Although Norris’ data consists of five waves, the survey tracks participants for
a total of only two years. Also, Norris’ study excludes those who do not remain lost
to follow-up.
68
2. Effects of Survey Design and Sampling Characteristics
Observation Plan and Changing Rates of Nonresponse over Time
In addition to individual characteristics that may influence a person’s
propensity to continue responding to the survey, there are also certain survey
attributes that may enhance the chances that a sample person will respond. For
example, there has been considerable debate over whether the length of time
between interviews influences response rates. Longer intervals decrease
respondent burden, and may allow more time to trace lost respondents (Kalton,
Kasprzyk, & McMillen, 1989). Alternately, some researchers have speculated that
higher overall response rates result from more frequent contact with sample
members. However, a comparison of response rates in the National Longitudinal
Survey of Labor Market Experience and the Survey of Income and Program
Participation (SIPP) found that even though the intervals between waves were
different, response rates in both surveys were stable over the years. This suggests
that neither frequency of contact nor length of survey lifetime affect rates of
nonresponse (Lillard, 1989). Although a common pattern of response rates shows
a dip at the first follow-up interview, then a leveling off (Lillard, 1989; Streib,
1982), this initial flux probably represents the weeding-out of those who are not
interested. In general, regardless of the length of time between interviews, most
researchers have found that the rate of nonresponse does not vary as a result of the
passage of time (Lepkowski et al., 1989; Lillard, 1989; Norris, 1985). Likewise,
69
except for the accumulation of nonresponse over time, there is apparently no effect
of the total number o f interviews.
Other Survey and Sampling Characteristics
One of the most potentially biasing sources of nonresponse may come from
lack o f interest in a survey topic. Downes (1952), for example, found that families
that did not have a member with a chronic illness were less likely to respond to a
follow-up survey on chronic illness (in Strieb, 1982). Lillard (1989) has
commented that following multiple households belonging to the same family in
studies such as the PSID increases response rates. Among older people, the
institutionalized population may have even higher rates of nonresponse than the
community dwelling population. Lower response rates may be expected among
persons who are living in institutions because they are more difficult to locate or
because of resistance from the institution or caregivers (Rodgers & Herzog, 1992).
3. Summary
Regardless of the definition of nonresponse, type of sample, or duration of
the study, there are several factors that are consistently implicated as prime
suspects in causing or covarying with nonresponse.
In general, those who are of lower socioeconomic status and in poorer
health are more likely to be non-respondents. Specific measures of these
constructs vary from study to study, but include race/ethnicity, educational
attainment, employment status, family income, types of dwelling structure, and
tenure type. Health status has been measured as mortality, institutionalization,
70
dependence on proxy, cognitive functioning, and morbidity, including self reported
health status and measures of functional health (usually ADL and IADL scales).
There are several other factors that are somewhat less clear predictors of
nonresponse. For example, household composition and family composition have
neither been consistent in their effects nor in their level of significance. For
example, marital status is sometimes significant, and sometimes not. Men are
usually found more likely to be non-respondents, although sex has also been found
to be insignificant. Living with others sometimes has a positive effect on
nonresponse, but sometimes the effect is negative. Although age is almost always
a predictor of the likelihood of nonresponse, it has been found to be insignificant in
some studies. The effects of age on the types of nonresponse are suggestive but not
conclusive. The inconsistency in effect of these variables may be because they are
either confounding or are confounded by some other variable.
There has been some work done on changing rates of attrition in
longitudinal panel surveys, but none of the work thus far has considered changes in
rates of attrition in samples of older adults. The only exception to this is a brief
note that presents descriptive statistics on percentages of nonresponse in the LSOA
(Chyba & Fitti 1992), which indicates that the rate of nonresponse increases over
the three follow-up waves. This finding contrasts with the analysis on younger
samples, which found rates of attrition to be constant.
Except for the work of Norris (1985), there has been no further
consideration of whether the characteristics of non-respondents vary over time.
71
And, although Norris’ work is an excellent starting point, the time frame for the
entire study is equivalent to one between-wave interval in the LSOA. This is not
sufficient to determine whether or not changes in non-respondents are a function of
historical time or survey time. While this cannot be addressed conclusively in one
or two studies, when research in this area begins to accumulate, patterns may
become evident.
C. Modeling Loss to Follow-up
There has been only limited consideration of how the causes, covariates and
rates of nonresponse may vary over the duration of panels of older samples
differentially affecting each wave. In particular, previous studies of wave
nonresponse in older samples have suffered from two important shortcomings:
First, researchers have typically chosen to consider a limited number of predictors,
usually basic demographic characteristics and perhaps a few other measures of
interest to the researcher. And, although some use multivariate techniques, most
consist of uncontrolled comparisons. This may account for the apparently
inconsistent effects on loss to follow-up of some of the characteristics reviewed in
the next section. Second, for the most part these studies consider only changes in
the sample between two points in time. Most panel surveys, however, consist of
more than two interviews. Therefore, it is important to assess the effect of loss to
follow-up over multiple interviews. In this section, a multivariate analysis
predicting nonresponse over the four waves of the LSOA is used to identify
72
characteristics of sample members associated with loss to follow-up due to factors
other than death.
1. Analytic Approach
The goal of this analysis is to account for the possibility that causes,
covariates, and rates o f nonresponse may vary over time, differentially affecting
each wave. Therefore, a method that allows the incorporation of time-varying
covariates is required. And, because the event of nonresponse takes place in the
lifetime of the survey and has the possibility of occurring at only three points in
time, the analysis requires a method suitable for handling coarsely measured
discrete time intervals and multiple ties in the event of interest.
A discrete time event history approach is chosen to estimate covariates of
nonresponse to take full advantage of information on timing of transitions to
nonresponse (Teachman, 1982). To accommodate the discrete-time observation of
transitions, a logistic regression model is chosen and adapted to approximate a
hazard rate (Allison 1984; Vuchinich, Teachman, & Crosby 1991; Yamaguchi
1991). In this case, the odds for the conditional probabilities of the event of
nonresponse occurring are modeled. This model has the form:
ln{X(r1 .;X)/[l-XfeX)]}=a,.+EAX*
where X(r,;X) is the conditional probability of becoming a nonrespondent at time t ,-
for the covariate vector X (X = X „ ...X t) and parameters b* (fc = l,...,I) and a, is the
log-odds for the baseline group (a^lnJX o^-V P-X o^]}). What is different from a
standard logistic regression, however, is the structure of the input data set. In this
73
approach, a pooled data set is created in which each observation gets a unique
record for each time interval (n = 10,758)2.
When the covariates are time-constant, this is a proportional odds model,
such that the odds of becoming a nonrespondent form a constant ratio with respect
to time among groups that are distinguished by the covariates. However, time-
varying covariates are included in most of the analyses. Therefore, among the
groups that are distinguished by the time-varying covariates, the odds of becoming
a nonrespondent varies with respect to time. The introduction of variables
indicating each time period allows us to determine whether there is an independent
effect of time (without specifying a specific functional form of time), and will
2 An analytic issue arises when the data structure is manipulated in this way
because there are now multiple observations for each individual. This introduces the
possibility of unobserved person-specific effects that will be correlated across the
observations for an individual. Tests of significance may be inflated (causing
associations to appear statistically significant when they are not), and estimates may
be biased as a result. (The problem of unobserved heterogeneity is also discussed in
Ch. IV, Section D2.)
Standard errors can be corrected, and the existence o f unobserved
heterogeneity may be tested for and identified, with random effects and fixed effects
models. By nature this study involves an unbalanced panel, and standard estimation
procedures in commercial software packages are not yet able to estimate such models
on this type of data. Allison (1994), describes the use of SAS PROC GLM (SAS
Institute Inc., 1989) with an optional ABSORB statement to estimate a fixed effects
model that can be used on data for whom the number of observations varies over
individuals. The ABSORB statement adjusts for the person-specific effects before
estimating the model with the predictor variables. A drawback o f this procedure is
that it does not produce estimates for predictive characteristics that do not change over
time. For the model specified in this chapter, results from this procedure indicate that
there is a significant person-specific effect. Estimates for those characteristics that
change over the course of the survey are similar both in their sizes and levels of
significance to those estimated using a regular logistic regression. Age, however, is
no longer an important predictor of nonresponse.
74
answer the question of whether the rate of nonresponse varies over time (strictly
speaking, it will answer the question of whether the odds of becoming a
nonrespondent vary over time).
2. Model and Measures
For the purpose of this study, a transition to nonresponse will be defined by
the first record o f non-interview status for any reason other than death.3 Because
the sample frame for the LSOA consisted o f respondents to the 1984 HIS, there are
no non-respondents for the first wave, thus eliminating the problem of left-
censored cases. Individuals can become nonrespondents at any wave after the
baseline interview (1984). The event of nonresponse takes place in the lifetime of
the survey and has the possibility of occurring at three points in time.
A model of loss-to-follow-up is posited based on previous findings
(reviewed above). It is expected that being older, male, African American, having
fewer years of education, not owning a home, currently residing in an institution,
reporting worse overall health, and reporting a greater number of functioning
limitations are associated with higher odds of nonresponse. Indicators of the
interval of observation are also included in order to test whether or not there is an
effect of time on the probability of nonresponse. An indicator for having another
3 Predictors and rates of attrition from death have been considered elsewhere.
However, it should not be assumed that all loss to follow-up conforms to the same
patterns as mortality (Baltes 1968, Siegler and Botwinick 1982, Manton 1988).
75
sample person in the household is included to indicate family involvement in the
survey.
To take advantage of fresh information collected at each interview, a pooled
data set is created in which each observation gets a unique record for each interval
(n = 10,756) and characteristics at the beginning of the interval are related to
response status at the end of the interval. This approach assumes that the error
terms for each observation (in this case each person-interval) are not correlated.
The record for each person-interval contains two kinds of independent
variables (Table IV-1). The first consist of those variables that were measured
only at baseline, and thus for analytic purposes, must remain the same at each
interval. These variables typically remain constant over adult life: sex, race, and
years of education. The indicators for having another sample person in the
household and self-rated health status also remain constant. The second type
consists of those variables that were collected or can be inferred at each interview:
age, home ownership, whether or not the respondent lives in an institution, general
health status, and functioning abilities. An indicator for interval of observation is
also attached to each person-interval record.
3. Results of a Multivariate Analysis of Loss to Follow-up
A series of four nested logistic regression models were run so that the
effects of including groups of predictor variables covering different characteristics
could be compared (Table IV-2). This approach clarifies inconsistencies in
findings among previous studies.
76
In the first model, a basic set of sociodemographic variables is included.
Being female and being older both increase the odds of becoming a nonrespondent,
while higher levels of education reduce the odds that a person will leave the
sample.
In the second model, measures of living arrangements and home-ownership
are introduced. The effect of sex becomes insignificant, and the odds for
becoming a nonrespondent are twice as high for those who are currently living in
an institution compared to those who are not living in an institution, and are lower
for home owners than for renters. Household composition is not significantly
related.
In the third model, two measures of health are included: self-reported health
status, and the number of functioning difficulties. As would be expected, those
who have poorer self reported health status have higher odds of being lost to
follow-up, as do those who have difficulty with a greater number of functioning
activities. And, with health status controlled, living alone becomes a significant
predictor of nonresponse.
In the fourth model, indicators of whether or not there are other sample
persons in the same household and the interval of observation are included. A
total of 1,861 of the 5,151 sample members at the baseline interview are living in
a household with another sample member; 921 households have multiple sample
77
members4. If a sample person lives in such a household, his or her odds of
nonresponse are over one third higher than for those who live with nonsample
persons5. Response status of those "other persons" is not controlled in this
analysis.
The coefficients of the time variables indicate whether (and how) the odds
of becoming a nonrespondent differ at the second and third intervals with reference
to the first. The significance of the dummy variable for the third interval indicates
that those persons responding at the third interview are more likely to be lost by
the fourth interval than respondents to the first interview were to be lost by the
second interview, all other characteristics constant.
Characteristics Affecting Probability of Nonresponse
Having identified which characteristics are associated with nonresponse in
this sample, the next question is: How can variation in these characteristics affect
the actual probability of becoming a nonrespondent? To explore this question, the
4 Multivariate analyses presented in this chapter, Chapter VI, and Chapter VII
include all sample persons in each household that are eligible for the analysis. This
introduces the possibility of biased estimates because of correlated errors across those
observations from the same household. To check that this has not had an effect on
the results, each analysis has been repeated on a sample consisting of one person
randomly selected from each household. Small changes in patterns of significance
result from the smaller samples in these analyses, as weaker predictors become
insignificant.
5 When the indicator for having another sample person in the household is
included in the model, the comparison group for "living alone" changes. It now
refers to those living with other nonsample persons.
78
mean probability of becoming a nonrespondent is estimated for selected categories
o f each o f the significant predictor variables (Table IV-3a).
For example, if the entire sample were age 70, and each individual retained
all other observed characteristics, the mean probability of becoming a
nonrespondent would be 7.5%. If the whole sample were 90 years old the
probability of becoming a nonrespondent would increase approximately three
percentage points. The effect of this 20 year increase in age is to increase the
probability o f nonresponse by 47%. Similarly, having another sample person in
the household results in a 30% increase in the probability of nonresponse to the
subsequent interview.
If the entire sample had difficulty with all thirteen o f the ADL and IADL
functions, the mean probability would increase 80% from .0866 (having no
difficulty with any ADLs) to .1551. This is the largest contrast among the
conditions listed.
The second column (a) of Table IV-3 indicates that the predicted mean
probabilities of loss for sample individuals who actually have the observed values
of the characteristics listed differ more widely than the probabilities in column a.
This increased difference is caused by the fact that those who are more or less
likely to be nonrespondents in one characteristic are likely to also have other
characteristics leading to similar outcomes. The only exception to this is in the
confounded relationship between living arrangements and having other
nonrespondents in the household.
79
4. Summary
Some results o f this analysis are consistent with previous research. Those
who are older, have lower educational attainment, are in worse health, and rent
their homes are more likely to become nonrespondents. However, no effect of
race, a characteristic that several previous studies found to predict nonresponse, is
found. Among those variables previously identified as ambiguous in their
relationship to nonresponse, sex becomes insignificant when measures of living
arrangement are included. And, when measures of health status are included,
living alone becomes a significant predictor of nonresponse. The effect of
currently living in an institution is also significant.
Some surveys collect information from multiple members of a single
household so that studies involving household behavior may be completed. In this
survey, having another sample person in the household increases the probability of
nonresponse considerably, relative to having another non-sample member in the
household. Exactly why this occurs is unclear. Those who have other sample
persons in the household are more likely to be older. If one household member
becomes ill, the other member may become the caregiver. In such a case both
sample persons might decide not to respond. Changes in the situation of the
household and the choice to not respond may occur after the last data were
collected and thus will not be accounted for in the analysis. Household response
behavior is clearly an area in which further research would be fruitful. The lesson
from this finding is not that survey designs including multiple sample members
80
from a single household should be avoided — this would result in a loss of valuable
information. Rather, awareness that such sample persons may be more prone to
loss to follow-up might call for special attention to retention of such sample
persons and possibly oversampling. Post-hoc analysis using such data may include
corrections for the disproportionate loss among these members.
Studies of nonresponse in panels with younger samples find that there is no
effect of the interval on the odds of becoming a first time nonrespondent. In this
sample o f older persons, however, incidence of first time nonresponse becomes
more likely as time passes.
Because most nonresponse interviews did not have a specific reason for
noninterview recorded, it was not possible to analyze nonresponse by reason. It is
important that future survey collection efforts monitor reason for nonresponse. If
nonresponse is occurring because of physical difficulties or anticipated respondent
burden, as might be inferred from these results, alternative or adapted interview
mechanisms could be used.
These findings suggest that wave nonresponse in this sample of older
persons is not random. Characteristics found to be associated with propensity to
become a nonrespondent may be used to develop targeting criteria for those
interested in retention of sample persons. And, for those using panel data with
older samples, the process of identifying salient characteristics and computing
probabilities of becoming a nonrespondent may be used to develop weights to
compensate for decrement in the sample.
D. Adjusting for Loss to Follow-up
1. Background: Correcting for Nonresponse
If loss to follow up occurs nonrandomly, that is, differentially selecting
persons with certain characteristics out of the sample, attention must be given to
correcting potential bias. There are three common approaches to correcting for
missing responses: discarding incomplete records, weighting, and imputation (Little
& Rubin, 1987).
The simplest approach is discarding incomplete data (for example, M or et
al., 1989; Rogers, Rogers, & Belanger, 1992). This approach is expedient but
should be accompanied by an analysis of the impact of excluding observations.
Some baseline analysis of characteristics related to the outcome of interest may be
used to determine whether excluding nonrespondents will result in a seriously
selected sample. One approach is to perform the analysis both on the selected
sample and on the sample including those who later become nonrespondents (see
for example Maddox & Clark, 1992). Similarly, Corder, Woodbury and Manton
(1992) apply the grade of membership approach to 28 variables indicating health,
functioning, and response status in parallel analyses for both the total eligible
sample and on a sample of only those who later become nonrespondents. The
resulting "pure type" profiles are similar for both samples, and the authors
conclude that no bias is introduced by the nonrespondents. Presumably, then,
these observations could be excluded from the analysis with minimal consequence.
82
Weighting is perhaps the most commonly used approach to unit nonresponse
adjustment. Two methods are commonly employed, cell adjustment and response
probability weights. The first approach uses cross-tabulations of the characteristics
affecting nonresponse to divide the sample into subgroups. Each cell is then
weighted with the inverse response probability. Thus, cells representing subgroups
of the sample population with characteristics associated with higher rates of
nonresponse will be "weighted up" to compensate for the differential loss. This
method is described in more detail in Rowland and Forthofer (1993).
Response probability weights work on the same principle as weighting class
adjustments. However, the weights are constructed from multivariate regressions
predicting probability of nonresponse. Therefore, whereas weighing class
adjustments provide weights for coarse groupings of subpopulations, response
probability weights can be as unique as the distribution of characteristics predicting
nonresponse (Iannachione, Milne, & Folson, 1991). Class based weighting
procedures and response probability weights have been found to produces quite
similar results (Lepkowski, Kalton, & Kaspryzk, 1989). While weighting is the
most widely used approach to correcting for nonresponse, and it will correct bias
on observed variables, it also tends to increase variances (Iannachione et al., 1991;
Kish, 1990; Little, 1986). It is, however, possible to adjust tests of significance
for this increased variance, just as it is done for sampling weights.
Imputation is similar to weighting, except that instead of giving respondents
additional weight to compensate for those lost with similar characteristics,
83
imputation actually replaces the missing data. The procedure is to locate other
observations with matching responses on items for which information exists, and
then allocate to the nonresponder the same values for the missing items as the
matched observations. Or, in the case where some waves of response are available
and an assumption of stability of characteristics is justified, an individual’s
response on earlier completed interviews may be used for later missed interviews
in a "carry-over" imputation procedure (Kalton & Miller, 1986). Although
weighting and imputation produce similar sample means (Kalton, 1990; Kalton &
Kish, 1981) imputing has the advantage of controlling both bias and variance
(Little, 1986). Imputation is commonly employed to replace missing data in both
the U.S. Census and Current Population Surveys. And, although imputation
procedures are usually reserved for nonresponse to single survey items, exceptions
exist. For example, in cases where there is evidence of a household but no
interview data has been collected, the Census may impute records for entire
households (Shryock & Seigel, 1976). Also, it has been suggested that imputation
might be fruitfully employed in combination with weighting for multiple wave
surveys in cases when a sample member who was once a respondent and becomes
a nonrespondent responds again at a later interview (Kalton, Lepkowski, & Lin,
1985).
2. Effect of Nonresponse Adjustments
All of the approaches described above assume that those who are lost to
follow up are similar on unobserved characteristics to continuing respondents that
84
have the same observed characteristics. This assumption may be untenable. It is
possible that there are unmeasured characteristics that influence an individual’s
response status. For example, it may be that some people do not like to respond to
surveys. If there are such unobserved characteristics influencing nonresponse, then
a correction based on observed characteristics will not eliminate the bias.
Unobserved characteristics that are associated with nonresponse, but not with the
outcome of interest in the substantive analysis (in this study, functional ability), or
any correlates of that outcome, would not affect the results o f that analysis. If
persons in worse health were less likely to respond to the survey than those in
better health, and health status was not included in the construction o f the
compensatory weights, then the weights would not adequately adjust for the loss of
the least healthy people in the sample. In that case, characteristics that are actually
associated with the worst health might appear insignificant because o f the
truncation of the health distribution. The same consequences would result if
correlates of health are neglected in this way. Functioning status, and all of the
characteristics that were hypothesized to be associated both with functioning status
and nonresponse were included in the model predicting nonresponse. It is
therefore unlikely that there are other unmeasured characteristics that would have
caused the sample to become selected on functioning status that would not be
corrected for in this analysis. (Refer also to footnote 2, page 74).
When descriptive univariate or bivariate statistics including prevalence and
transition rates are being reported for a sample with nonrandom nonresponse,
85
correction may be required. However, when a multivariate model is fully
specified in a substantive analysis and there are no unobserved characteristics
contributing to the propensity to be lost to follow-up, then the model itself should
control for the distribution of characteristics in the sample. The only way to
determine how much difference an adjustment for nonresponse will make is by
making the adjustment and then comparing corrected and uncorrected analyses. A
number of studies have found that correction did not alter the interpretation of the
results (for example, Kalton et al., 1989; Rowland & Forthofer, 1993).
3. Analytic Approach to Compensating for Nonresponse
The probabilities of nonresponse calculated in the previous section are now
used to construct weights to compensate for nonresponse over the three survey
intervals in the LSOA. The effect of this strategy is compared with an uncorrected
sample, and with the sample using only weights provided to correct for design
effects.
Constructing Weights
Weights for the data are constructed based on the probabilities of
nonresponse for characteristics found to be salient predictors of nonresponse in the
previous analysis. The probabilities of nonresponse are estimated within each
interval of observation. While this approach is not new, response probability
weighting typically makes use of only first wave information (Kalton & Miller,
1986). In other words, even if a sample member responds to more than one
interview, only their responses at the initial interview are used to correct for their
86
future nonresponse. Clearly this could result in considerable waste of observed
data. In this approach, each sample member is retained as long as he or she has
not missed an interview. Therefore, if a person does not become a nonrespondent
until 1990, three waves of original data are retained. The remaining sample is
then weighted up to compensate for this person’s loss in the final wave. This
strategy allows considerably more information to be included. This is particularly
important in longitudinal surveys that collect information on characteristics that are
not expected to be stable, and for which precision in estimating transitions is
critical. Some loss of information may be expected because reinterviews of
persons who were previously nonrespondents are not taken into account.
However, because only 47 persons in this study have a pattern of response that
would contribute in this way (see Table III-3, pattern NIN), the impact is probably
quite small.
For each interval (e.g. 1986-1988) for all sample members responding at
the beginning of the interval, an indicator of response status is regressed on
variables that were significant in the pooled analysis. These probabilities are then
applied to the sample weight for design effect to correct for nonresponse.
Because the LSOA is a multistage cluster sample, the LSOA sample weight
is provided to adjust by geographic area for baseline nonresponse and the
subsampling of certain populations. The weight adjusts the baseline sample so that
it is representative of the noninstitutionalized population that is 70 years of age and
over in 1984 by sex, race and age. This sample design weight is used for the
87
baseline data, and then modified cumulatively at each successive wave to
compensate for nonresponse. Thus, each sample person has a wave-specific weight
that corrects for both the subsampling and that person’s propensity to be lost to
follow up. When these weights are applied, the contribution of those persons who
actually respond are inflated to also represent those who did not respond, based on
characteristics included in the computation of the probabilities. Thus, the weights
increase the sample size to approximately what the sample would have been if the
only source of attrition had been death, and they alter the distribution of the
characteristics so that they take into account the propensity of some groups to
become nonrespondents more than others.
Effect of W eighting for Nonresponse
Table IV-4 compares the effect of weighting only with the normalized
LSOA weight to correct for sample design with the correction weight for
nonresponse on the distribution of the number of ADLs and IADLs across the four
interview waves. (In 1984 the nonresponse weight is equivalent to the LSOA
weight because there is no baseline nonresponse to account for). The general
pattern is that the LSOA weights alone make the sample a bit healthier at each
wave compared to the unweighted sample, adding a few more people to the zero
difficulties category. Alternately, since those in poor physical functioning are less
likely to respond, the nonresponse weights compensate for that tendency, and thus
move a small proportion out of the zero difficulties category and into the categories
indicating difficulty with one or more activities. Still, even though the
nonresponse weights move the sample toward worse health, they do not return to
the levels of the unweighted data. In 1986 the effect of the nonresponse weights
(compared to the data weighted with LSOA sample weights) is to move about .2
percentage points of those in the zero difficulties categories to a category of having
difficulty with one or more activity, for both ADL and IADL. In 1988 and 1990
the nonresponse weights decrease the zero difficulties category by roughly .8
percentage points for both ADL and IADL, and the effect is distributed among the
other categories. In 1990 there is a subtle trend for the target of this redistribution
to be among categories of more disability. In fact, for IADL in 1990, the
nonresponse weights move people out of both the zero difficulties category and the
one difficulty category.
In this table, it can be seen that by 1990 the actual sample responding to
ADL questions had diminished to 2,635. The nonresponse weights have increased
the sample responding to this question by 641 persons, making the sample just 18
shy of what the total sample would have been in 1990 if the only source of
nonresponse had been death.
Discussion
For those using panel data with older samples, the process of identifying
salient characteristics and computing probabilities of becoming a nonrespondent
may be used to develop weights to compensate for decrement in the sample. The
approach suggested here allows for optimal use of each sample members’
contribution to the survey by estimating response probabilities within each interval
89
and applying these to form a cumulative weight. Such a method is a step toward
restoring both sample size and redistributing characteristics to maintain
representativeness over time.
E. Summary and Discussion
All longitudinal panel studies are susceptible to selection because sample
persons either chose not to respond, become unable to respond, or because sample
persons relocate and cannot be found. This study addresses the extent of the loss,
the degree to which the loss is associated with certain hypothesized characteristics,
and it evaluates the effect of a weighting scheme to compensate for nonrandom
nonresponse.
An event history approach is employed to ascertain whether some groups
are more likely than others to become nonrespondents and whether the rate of
attrition varies over time. The approach taken in this study addresses several
important shortcomings of previous analyses of nonresponse: It investigates a
variety of characteristics hypothesized to be associated with nonresponse within a
multivariate framework, and it incorporates a dynamic design to take advantage of
time-varying covariates in the analysis of loss to follow-up over multiple
interviews.
To some extent, who will become a nonrespondent can be predicted. For
example, persons who are older, have lower levels of education, are in worse
health, are living in an institution, and have another sample person in the
household are more likely to become nonrespondents. This last characteristic is a
90
finding that has not been previously reported and has important implications for
survey design and protocol. Because the inclusion of multiple sample persons from
a given household is a feature that significantly enrichens a data source, special
attention must be paid to retaining these survey participants. It is a matter of
speculation, however, why such persons are more likely to be lost. Although such
factors as the relative older age of couples selected from the same household and
some aspects of failing health have been controlled, it is possible that either
increased frailty or caregiving demands of one member for the other may reduce
survey participation. Similarly, even though this effect exists regardless of the
"other" sample person’s response status, a decision dynamic may exist in
households with multiple sample persons that works against continued participation.
This would be a fruitful area for future research.
Isolating specific characteristics such as these facilitates several
compensatory approaches. First, on the level of survey design and protocol, it
allows subgroups to be targeted for either oversampling or for more intensive
strategies for retaining those persons in the sample. Second, for those using panel
data, this strategy for identifying characteristics associated with nonresponse is
especially useful because it maximizes the information from time-varying
covariates. This allows a more precise specification of those characteristics that
are associated with nonresponse. Importantly, too, it has permitted the
examination of the effect of duration of time spent in the study on propensity to
become a nonrespondent, net of other factors. Also, precise and complete
identification of characteristics associated with nonresponse is important for
constructing appropriate compensatory weights.
Most statistical evaluations of the selectivity effects of nonresponse do find
a significant nonrandom component. However, whether the selection will
introduce bias into the results can only be assessed by comparing corrected and
uncorrected results. In this study, a strategy for compensating for nonresponse is
proposed that uses a cumulative weighting scheme. This approach is different
from common approaches that weight for wave nonresponse occurring over the
course of the survey based on characteristics of eventual nonrespondents at the
inception of the survey. Such a strategy does not take into account the possibility
of change in characteristics between that first interview and the time at which the
sample member becomes a nonrespondent; this constitutes a considerable waste of
information. Because the approach presented here retains sample persons in the
analysis until the wave in which they do not respond for the first time,
nonresponders contribute the maximum amount of information. Because this
approach takes into account the most recently collected information on the
nonrespondents’ characteristics, the weights are as precise as possible within the
constraints of the data.
The data presented here suggest that wave nonresponse in this sample of
older persons is not random. Characteristics found to be associated with
propensity to become a nonrespondent may be used to develop targeting criteria for
those interested in retention of sample persons. And, for those using panel data
92
with older samples, the process of identifying salient characteristics and computing
probabilities of becoming a nonrespondent may be used to develop weights to
assess the effect o f nonresponse and to compensate for selective decrement in the
sample. The approach suggested here allows for optimal use of each sample
members’ contribution to the survey by estimating response probabilities within
each interval and applying these to form a cumulative weight. Such a method is a
step toward restoring sample size and redistributing characteristics to maintain
representativeness over time. Outcomes adjusted with these weights are compared
with unadjusted numbers, and the insignificant differences lead to the conclusion
that even though nonresponse did not occur randomly, it does not result in serious
bias in these data, for this outcome of interest.
93
Table IV-1: Variable Definitions and Coding by Observation Plan
Measures collected only at baseline
Sex
Race
Years of Education
Health Status
Other Sample Person in Household
(l=female, 0=men)
(l=black, 0=all others)
(in actual years: 0-18+)
(l=excellent, 5=poor)
(l=yes, 0=no)
Measures collected at each interview or chanae can be inferred
Age (years reported)
Non-response (l=yes, 0=no)
Living Alone (l=yes, 0=no)
Owns Home (l=yes, 0=no)
Living in institution at interview (l=yes, 0=no)
Number ADL“ and IADLb difficulties (0-13)
Duration of time in survey (l=yes, 0=no)
* "Activities of Daily Living" is a commonly used scale to measure
functioning ability in the realm of basic personal care needs. The
standard seven item scale includes indicators of whether if, for
health reasons, an individual has difficulty with the following items:
bathing or showering; dressing; eating; getting in or out of bed or
chair; getting outside; walking across a small room; and using the
toilet.
b "Instrumental Activities of Daily Living" is a commonly used scale
to measure functioning ability in the realm independent household
maintenance. The standard six item scale includes indicators of
whether if, for health reasons an individual has difficulty with the
following items: preparing one's own meals; shopping for personal
items; managing own money; using the telephone; doing light housework;
and doing heavy housework.
94
Table IV-2; Odds Ratios for Significant Predictors of Nonresponse
Model 1 Model 2 Model 3 Model 4
Odds Ratios Odds Ratios Odds Ratios Odds Ratios
Intercept 0.014“ 0.025“ 0.023“ 0.022“
Sex 1.189*
Race
Age 1.031“ 1.027“ 1.023“ 1.022“
Yrs Educ 0.945“ 0.948” 0.955“ 0.951“
Lives Alone 1.180’ 1.383“
Lives in Inst 2.108* 1.945* 1.946*
Owns Home 0.604“ 0.616“ 0.607"
Health Status 1.080* 1.079*
Functioning 1.030* 1.032*
Other Sample Person 1.354’
Interval 2
Interval 3 1.223*
-2 Log L 6148.962 6092.729 6075.865 6060.903
covariates m2 v ml m3 v m2 m4 v m3
Chi-sq 78.627 56.233 16.864 14.963
df 4 3 2 3
P
.0001 .0001 .001 .01
response categories 1 = 930 1 = 930 1 = 930 1 = 930
0 = 9,270 0 = 9,270 0 = 9,270 0 = 9,270
p< .05
* * p<.001
VO
Table IV-3: Predicted Mean Probability of Becoming a Nonrespondent for
(a) Specific Values of Significant Independent Variables with All
Other Characteristics as Observed
(b) Sample Members Observed to have Specific Values of Significant
Variable Value (a) Mean p fbt Mean p
Age
Age 70
Age 90
.0755
.1112
.0819
.0950
Education
8 Years School
16 Years School
.0977
.0681
.1005
.0661
Living Arrangement
Lives with Others
Lives Alone
.0806
.1076
.0830
.1032
Tenure Type
Does Not Own
Owns Home
.1243
.0798
.1319
.0783
Living in an Institution
No
Yes
.0904
.1602
.0903
.1910
Self-Rated Health Status
Poor Health
Excellent Health
.1049
.0800
.1237
.0738
Functioning Health (ADL/IADL)
No Difficulties
Diff. w/ 6 functions
Diff. w/ 13 functions
.0866
.1025
.1551
.0778
.1252
.1553
Other Sample Persons in Household
No .0840
Yes .1098
.0927
.0881
Respondent at 3rd Nave
No
Yes
.0867
.1037
.0862
.1056
96
Table IV-4: Comparison of Frequencies for ADLs and lADLs in Each Wave Without Weights, With Only the LSOA Sample Correction Weight, and with Weights
C onrtnjctedtoC o2ertforProbabijitvrfN onres|>onge_
Unweighted
ADL
LSOA Weights Nonres. Weights Unweighted
IADL
LSOA Weights Nonres. Weights
1984
1986
1988
1990
VO
N % M % N % N
% .
N % M %
0 3637 71.0 3770 73.6
..
3341 65.3 3511 68.6
1 526 10.3 507 9.9 829 16.2 800 15.6
2 319 6.2 291 5.7 308 6.0 271 5.3
3 199 3.9 176 3.4 n. a. 170 3.3 150 2.9 n.a.
4 151 2.9 129 2.5 191 3.7 158 3.1
5 104 2.0 95 1.9 135 2.6 110 2.2
6 114 2.2 94 1.8 - -
. .
141 2.8 116 2.3 - - --
7 69 1.3 60 1.2
— — — —
5119 100% 5122 100% 5115 100% 5119 100%
0 2387 58.7 2548 61.5 2740 61.2 1991 51.0 2159 53.8 231? 53.6
1 579 14.2 580 14.0 628 14.0 897 23.0 921 23.0 955 23.0
2 332 8.2 319 7.7 347 7.7 313 8.0 302 7.5 330 7.6
3 219 5.4 203 4.9 221 4.9 192 4.9 179 4.5 195 4.5
4 157 3.9 149 3.6 163 3.6 165 4.2 146 3.6 160 3.7
5 107 2.6 97 2.3 106 2.4 145 3.7 125 3.1 136 3.1
6 171 4.2 153 3.7 167 3.7 203 5.2 179 4.5 195 4.5
7 117 2.9 95 2.3 104 2.3
— — — — — —
4069 100% 4143 100% 4475 100% 3906 100% 4011 100% 4326 100%
0 1837 56.0 2028 59.3 2260 58.6 1745 53.5 1897 55.7 2102 54.8
1 494 15.1 506 14.8 576 15.0 696 21.3 743 21.8 843 22.0
2 260 7.9 251 7.4 280 7.3 256 7.9 249 7.3 184 7.5
3 182 5.5 180 5.3 211 5.5 163 5.0 156 4.6 181 4.8
4 128 3.9 110 3.2 126 3.3 124 3.8 113 3.3 128 3.3
5 97 3.0 93 2.7 108 2.8 122 3.7 109 3.2 130 3.4
6 155 4.7 136 4.0 157 4.1 155 4.8 137 4.0 161 4.2
7 127 3.9 116 3.4 136 3.5
— — — — — —
3280 100% 3420 100% 3856 100% 3261 100% 3404 100% 3836 100%
0 1398 53.1 1567 56.1 1812 55.3 1398 53.1 1544 55.4 1791 54.7
1 407 15.4 424 15.2 499 15.2 540 20.5 583 20.9 681 20.8
2 229 8.7 243 8.7 289 8.8 196 7.4 196 7.0 238 7.3
3 140 5.3 137 4.9 162 5.0 139 5.3 129 4.6 154 4.7
4 108 4.1 101 3.6 126 3.8 108 4.1 106 3.8 133 4.1
5 87 3.3 81 2.9 132 3.1 94 3.6 86 3.1 109 3.3
6 129 4.9 117 4.2 133 4.0 156 5.9 142 5.1 166 5.1
7 137 5.2 123 4.4 154 4.7
— — — — — —
2635 100% 2792 100% 3276 100% 2631 100% 2786 100% 3272 100%
V. USING ITEMS FROM ADL AND IADL SCALES TO MEASURE
CHANGE IN FUNCTIONING
A. Introduction
The outcome of interest in this study is change in ability in Activities of
Daily Living and Instrumental Activities of Daily Living. Items on the Activities
o f Daily Living (ADL) scale are thought to identify the most severely disabled
individuals in the community. These measures include whether or not an
individual has difficulty bathing, dressing, eating, toileting, walking, and getting in
or out of bed or chair. Instrumental Activities of Daily Living (IADL) are
generally thought to measure the ability to live independently in the community.
These measures include the ability to shop for oneself, prepare meals, manage
money and use the telephone. Although ADLs and IADLs are commonly used
measures of functioning ability, autonomy, and need for assistance, different
formats of the questions capture different components of ability, and there is
considerable controversy over how to combine the individual items from the scales.
And, although a number of previous studies have investigated these issues, in every
case they focus on the use of the items to measure status, not change. The goal of
this chapter is to determine the most appropriate way to combine the items from
the ADL and IADL scales for use as a measure of disability in this study of change
in functioning. The following four questions identify the issues associated with the
use of these measures and will guide the investigations in this chapter:
98
(1) How many constructs or underlying dimensions of ability are
measured by the 12 ADL and IADL items? Or, do the ADL items
and the IADL items measure the same thing?
(2) Do ADLs or IADLs form a Guttman type hierarchy, as they are
often assumed to do? Do people lose their ability to do the
individual tasks on the ADL or IADL scale in a particular order?
(3) Is the difference between ADLs and IADLs one of degree (are
the scales ordinally associated) or kind (or are they categorically
different)?
(4) Do variations in how ability in ADL and IADL questions are
asked yield different rates of change (incidence of disability) in
functioning over time?
The answers to these questions clearly have important implications for usage (e.g.,
which question format should be used, and how items from the scales and each
scale should be combined) as well as interpretation. These are important practical
issues for this study, and are relevant to other researchers deciding how to
construct an outcome variable with ADL and IADL items. The issues addressed
here have broad ramifications for the wider research community as well, however.
There is no generally agreed upon method for aggregating ADL and IADL items —
indeed, there are few cases when researchers have used identical schemes. This,
as well as differences in how the ADL and IADL questions are asked even on
major national surveys has seriously compromised the comparability of research on
99
functioning health and aging. As more research is done on the differences in
question wording as well as the other measurement properties of the ADL and
IADL items, perhaps researchers can begin to move toward more directly
comparable techniques. Or, at least, knowing how approaches to using the ADL
and IADL items differ will allow post hoc adjustments to account for differences
between surveys.
In this chapter, several issues regarding the measurement properties and
usage of items from the ADL and IADL scales are evaluated. These issues include
the dimensionality and scalability of the individual items, and differences in
question format on the representation of change in functioning. Results from this
evaluation are used to direct the construction of the dependent variable used in
subsequent analyses in this study. This introduction proceeds with a more detailed
discussion of the issues that will be addressed, a description of the survey items
that will be used in the analyses, and an overview of the chapter.
1. The Issues: An Introduction
Although the substance of the ADL and IADL items seems relatively
straight-forward, a variety of issues exist regarding their use. While almost all
surveys of elderly health contain questions on ADL and IADL, there is no
generally agreed upon method of using these items to measure health in the cross-
section. Much discussion has centered around the relative validity and usefulness
of treating items from ADL and IADL scales as hierarchically related (as a
Guttman scale) or as falling along some latent constructs not necessarily coinciding
100
with the division between the scales themselves (such as arise from a factor
analysis or a grade of membership approach). In practice, most researchers default
to a summative scale, which assumes that all items are o f equal weight and reflect
the same dimension of functioning.
Although there is an unresolved debate in the literature regarding the
dimensionality and scalability of ADL and IADL items, the entire debate has
centered on status measures: measures taken at one point in time, and the
characteristics of the different approaches to combining the measures in the cross-
section. In this chapter, the basic dimensional and scaling properties of the sets of
measures that are now commonly included in ADL and IADL scales are
evaluated — both as status measures (that is, as they measure functioning status at a
point in time), and as measures of change in functioning between two time points.
In assessing the dimensionality and scalability of the ADLs and IADLs,
results from three forms of response categories used in collecting these measures
are compared: responses to questions of whether or not the respondent had
difficulty with the activity; whether or not the respondent was unable to perform
the activity; and whether or not the respondent received help with the activity. All
three of these variations are commonly used ways of coding ability or dependency
in the older population. Finally, the difference in the three question formats on the
representation of change is evaluated.
101
2. Description of Self-report Functioning Items in the LSOA
In the LSOA, three forms of the basic ADL and IADL questions are
available. First, each respondent is asked whether or not he or she has difficulty
with the item in question. For both ADL and IADL items, the interview probes
whether or not the this is because of a health or physical problem, and for ADL
items, the question is qualified to refer to ability "by yourself and without using
special equipment." Next, if the person does have difficulty they are asked how
much difficulty they have. Responses to this question are recorded as "some," "a
lot," and "unable." Also, for each activity, those who report difficulty are asked
whether or not they receive help with that activity. Thus, dichotomous measures
indicating difficulty, disability, and receipt of help are available, as is an ordinal
measure ranging from no difficulty to inability for each of the ADL and IADL
items.
3. Organization of the Chapter
This chapter is organized in two sections to address the issues introduced
above. The first section is concerned with approaches to combining items from the
ADL and IADL scales and includes an investigation of two aspects of the
measurement properties of those items (1) dimensionality and (2) scalability.
These investigations compare results using each of the three question formats
described above. In the next section, differences in how each of these question
formats reflect rates of change are evaluated. This section also includes a
discussion of the potential of response bias with ADL and IADL items and the
102
contribution of variations in question wording to this potential. The results from
this investigation inform the construction o f the dependent variable used in
subsequent analyses. The relevant results and rationale are summarized and
discussed in the final section.
B. Competing Approaches to Combining Items from ADL and IADL Scales
1. Introduction
The overriding issue throughout all studies of how to combine items from
ADL and IADL scales is whether they are capturing one, or more than one,
dimension of health. Embedded in this question is the question of whether items
within each (or both) of the scales are hierarchically associated.
Scalability
Although early research found that on a particular sample, selected IADLs
met all of the formal requirements of a Guttman scale (Rosow & Breslau, 1966),
items have been added to the scale and samples have become more representative,
so that thirty years later the same basic questions are still being investigated.
Results supporting the hierarchical approach have been presented by Spector, Katz,
Murphy, and Fulton (1987) who find that selected items from ADL and IADL
scales are both hierarchical and discriminant. These authors conclude from these
findings that it is legitimate to combine items from the scales, as is typically done.
However, many of the Guttman scale analyses use a limited set of items, and the
reason for this would appear to be that not all items on the full scales conform to a
Guttman hierarchy. Both Laforge, Spector and Sternberg (1992) and Pinsky,
103
Leaverton, and Stokes (1987) had to eliminate observations from their analyses
because the experiences of those sample members did not conform to the structure
imposed by the scale.
Dimensionality
Kempen and Suurmeijer (1990) argue that ADL and IADL scales are not
only hierarchical but that they do indeed form one dimension. In a complementary
study, Siu et al. (1993) find that when entered into a factor analysis with a variety
o f other self report and performance based measures, ADL and IADL items do
load together. Alternately, Wolinsky and Johnson (1991) have argued that ADL
and IADL items actually reflect three dimensions of functioning. Wolinsky
identifies the dimensions resulting from the original ADL and IADL items as
"basic ADLs", which consist of five ADL functions; "household ADLs", which
include four IADLs; and "advanced ADLs" (managing money, using the phone,
and eating) indicating cognitive capacity.
2. Dimensionality of Items on ADL and IADL Scales
(1) How many constructs or underlying dimensions of ability are
measured by the 12 ADL and IADL items? Or, do the ADL items
and the IADL items measure the same thing?
While almost all surveys of elderly health contain questions on ADL and
IADL functions, there is no generally agreed upon method of using these items.
Much discussion has centered around the relative validity and usefulness of treating
items from ADL and IADL scales as hierarchically related (as a Guttman scale) or
104
as falling along some latent constructs not necessarily coinciding with the division
between the scales themselves.
Although practice tends to reflect confidence that the items used in ADL
and IADL scales measure unique and distinct constructs, there have been
differences of opinion. For example, Kempen and Suurmeijer (1990) found that
ADL and IADL scales are hierarchical and scale along a single dimension; while
Wolinsky and Johnson (1991) have suggested that the 12 items scale along three
distinct dimensions. Some of this difference is probably attributable to using
different forms of the questions (Kempen and Suurmeijer use a three response
category version of levels of dependency, and Wolinsky and Johnson use
dichotomous measures indicating having difficulty), and some of the differences
probably also arise because they use somewhat different sets of measures, but the
deciding factor may be that Kempen and Suurmeijer use exploratory factor
analysis, and Wolinsky and Johnson use a hybrid approach, including both
exploratory and confirmatory factor analyses.
The results in Wolinksy and Johnson (1991) that resulted in the unusual
three factor solution were presented as findings from an exploratory analysis, and
thus as reflecting the "underlying structure" of ADL and IADL items. However,
those results were, in fact, obtained by theoretically driven manipulation of the
data. In their analysis, the researchers decided a priori that eating, using the
telephone and managing money represent a different construct from the other more
clearly physical tasks. They suggest that these three items reflect a cognitive
105
dimension of functioning. These three items are submitted alone to a factor
analysis and are found to load together. The remaining items are submitted to a
factor analysis and they are found to load along two dimensions — ADLs on one,
and IADLs on the other.
Because each study has used slightly different items in the ADL and IADL
scales, and they have been coded differently (dichotomous indicators of having
difficulty or ordinal measures of level of difficulty), it is difficult to directly
compare the results. Also, in the case of Wolinsky’s study, although the LSOA is
used and the measures used in that study could be used in this study, it is
important to ascertain how those three questionable items will behave in an
exploratory analysis. It is expected that ADL and IADL items will fall along two
factors corresponding with those scales. However, in this study, the focus is on
change. After verifying the structure of ADL and IADL items as measures of
status, the structure of change in ADL and IADL ability needs to be examined.
Two sets of factor analyses follow, one on status scores and one on change scores.
Sample
Because the issue of interest in this analysis is how measures of functioning
perform over time, the sample has been selected to exclude members who do not
respond to one or more of the interviews. Also, because data are collected or
coded inconsistently for those in institutions, those who are ever in an institution at
the time of an interview are excluded. Because the sample has been selected in
this way it is possible to compare various approaches to measuring functioning
106
over time on the same group of people as they age over six years. The results, of
course, are not representative of the U.S. population, but o f those persons 70 years
of age and older living in the community in 1984 who survive through 1990.
Basic demographic characteristics for the selected sample were compared to the
full sample in the previous chapter (Table IV-2). An evaluation of correcting for
nonresponse on the prevalence (Chapter III) and incidence (Mihelic & Crimmins,
1995) of ADL and IADL disability in this sample suggest that nonresponse
introduces negligible changes in the distribution of disability, and is therefore
expected to have an inconsequential effect on the results presented here.
Analytic Approach
There are two issues embedded in the question of dimensionality: First,
what are the underlying construct(s) that the items in question measure? Previous
research (reviewed above) has indicated that items from the ADL and IADL scales
might lie along one dimension, and thus be indicators of one underlying construct;
that they fall along two dimensions corresponding to the scales themselves; and
that they fall along three dimensions. It is important to know the dimensionality
underlying these items for two reasons: it has implications for how the items
should be combined in a summary index and also how the items themselves and
any summary measure should be interpreted.
The second question is whether the dimensionality of the items in the ADL
and IADL scales are consistent over time. This issue is sometimes referred to as
"factorial invariance." Factorial invariance is important because it indicates that
107
the items continue to reflect the same construct over time. This question is
typically posed when the same measures are being used on different populations
(especially samples of different ages). The present study is somewhat different
because it involves a panel sample — the same people at different ages. However,
a test o f factorial invariance will provide an indication of whether the items from
these scales are measuring the same thing at each wave in this sample (as the
sample ages and becomes, on average, more impaired).
Cunningham (1989) reviews various criteria that have been used to test for
factorial invariance. These tests range from the stringent metric requirements of
identical factor loadings, to less restrictive configural invariance, which in its most
lenient case requires simply that items load together to form the same factors
consistently over times or samples. A variety of arguments justifying the less
stringent criteria exist, including imperfections in sampling and measurement that
lead to violations of the assumptions necessary to achieve identical factor loadings,
and the substantive requirements of the research. These justifications apply to the
present analysis.
The objective of this analysis is to determine the number and composition
of dimensions reflected in the ADL and IADL items, and to ascertain whether or
not these dimensions are reliable over time. To address these questions, three
versions of the six ADLs and six IADLs are submitted to exploratory factor
analysis. SAS PROC FACTOR (SAS Institute Inc., 1989) is used to extract the
factors (using the principle component method), and oblique rotation is used
108
because the dimensions are assumed to be correlated. For simplicity, the loadings
in the tables are multiplied by 100 and rounded to the nearest integer. There is a
theoretical requirement that the original items used in factor analysis be continuous
(i.e. measured at an interval level). This requirement is necessary only for precise
studies concerned with the metric of the loadings. The goal of this study is simply
to reveal the underlying structure of the dimensionality of the ADL and IADL
scales and to look for reliability in this structure over time. For this sort of
heuristic endeavor, there are no serious consequences of violating that assumption
(Kim & Muller, 1978).
Factor analysis produces linear combinations (weighted sums) of the
original variables. The first factor is derived in such a way that its explanatory
value for the set o f original variables is maximized. The coefficients are
determined so the factor will have a mean of zero and a standard deviation of 1.
This construction ensures that there will be a unique set of solutions. The second
and subsequent factors are determined so that they each, in turn, maximize the
amount of variance they explain in the original variables (of the remaining variance
unaccounted for by the first and other preceding factors).
The order of the factors is determined (by construction) by the amount of
variance in the original items they explain. In other words, the first factor
explains the largest proportion of variance in the items, then the second, and so on.
When a group of items loads on one factor those items are considered to be
indicators of the underlying construct represented by that factor. The order of the
109
factor (factor 1, factor 2) does not have any bearing on the dimensionality of the
item group. Similarly, the factors may switch order. Again, this does not have
any bearing on the dimensionality of the items represented in those factors, but it
does reflect the relative explanatory value of the factors.
Dimensionality of Status Scores for Items on ADL and IADL Scales
Factor analyses are performed for the three forms of the basic ADL and
IADL items. Table V -l presents rounded factor loadings for items indicating
whether or not the person has difficulty with six ADLs and six IADLs. These
results suggest fairly stable dimensionality of the ADL and IADL items with the
exception o f the eating and heavy housework items. By 1990 the items have
aligned as expected (ADL items loading on one factor and IADL items loading on
another factor), except that having difficulty walking has shifted allegiance, and is
now loading with IADL tasks.
Table V-2 shows the results of the same analysis using indicators of
disability for each of the 12 items. In this analysis, except for the idiosyncratic
behavior of bathing in 1986, the ADL and IADL items load cleanly as two factors
corresponding to the respective scales. In Table V-3, rounded factor loadings for
items indicating receipt of help are presented. Again, these items load cleanly as
two factors corresponding to the ADL and IADL scales.
Dimensionality of Change Scores for Items on ADL and IADL Scales
The analysis thus far has used measures of functioning status at each
interview. But it is the structure of change that is of primary interest. Therefore,
110
the same analysis is repeated on items that indicate whether or not functioning
changed over an interval. Change scores for three sets of analyses indicate
movement from being able to being unable, from having no difficulty to having
difficulty, and from not receiving help to receiving help at each of the three
intervals (1984-1986, 1986-1988, 1988-1990). Tables V-4, V-5, V-6 show the
results of this analysis. Similar to the results of the factor analysis on status
scores, the receipt of help items (Table V-6) fall cleanly along two dimensions,
while both disability (Table V-5) and difficulty items (Table V-4) load less neatly.
While there is some shifting of items back and forth on the have difficulty factors
over the three intervals, after the first interval the disability items align on the two
expected factors (again, with the exception of disability bathing, now in the third
interval). The dispersion of the disability items over three factors in the first
interval may be the result of instability due to very small frequencies of persons
making transitions to disability in the first interval.
Discussion
Although Wolinsky and Johnson (1991), also using the LSOA, have decided
on largely theoretical grounds that ADL and IADL items measure three dimensions
of functioning, the results presented here from an exploratory analysis provide no
evidence to suggest other than two underlying dimensions: one for the ADL scale
and one for the IADL scale. While there are a few variations, the pattern
overwhelmingly suggests that indeed ADL and IADL items measure two distinct
dimensions of functioning whether status scores or change scores are used; the
111
variance in the 12 functions is largely captured in the two factors. Statistical
criteria would suggest that an additional factor is not missing. The structure is
robust across interviews as the sample ages, and there is little variation due to the
type o f question. The results are the same for a parallel analysis using the sample
not selected for response at all interviews.
On the other hand, among the three question types, the structure is more
cleanly captured by either the receipt of help or the disability items. All other
things equal, the have difficulty items might be expected to give the clearer picture
of the underlying dimensions because they are distributed more normally than the
responses on the other two question formats. They do not, however, and it may be
that their failure to load as neatly as the items from the other two question formats
occurs because of reliability problems with the "have difficulty" questions.
Response to questions on whether or not a respondent "has difficulty" with a task
item may be more susceptible to random fluctuation than response to questions
about whether or not the respondent is unable or is receiving assistance.
3. Scaling Properties of ADL and IADL Items
The Relationship Among Items on the ADL Scale and the Relationship Among the
Items on the IADL Scale
Having established that the items in the ADL scale reflect one dimension of
functioning, and the items in the IADL scale reflect another dimension of
functioning, the next questions concern the precise relationships of the items within
112
the scales among themselves, and of the two scales to each other. The specific
questions to be addressed are:
(2) Do ADLs or IADLs form a Guttman type hierarchy, as they are often
assumed to do? Do people lose their ability to do the individual tasks on
the ADL or IADL scale in a particular order?
(3) Is the difference between ADLs and IADLs one of degree (are
the scales ordinally associated) or kind (or are they categorically
different)?
The first question, whether or not items within the ADL scale form a
Guttman-type hierarchy, has been nicely addressed by Lazaridis, Rudberg, Fum er,
& Cassel (1994). These researchers found that of the 360 possible orderings of the
ADL items, 103 satisfied the minimum criteria for a Guttman scale, and that four
are at least as good as the traditional Katz hierarchy (bathing, dressing, toileting,
transferring, followed by the other tasks). In other words, people do not lose their
ability to perform this set of ADLs in a specific order, as has commonly been
assumed.
A test of the Guttman properties of six IADL items for four years were
tried for a number of likely orderings. The test for a Guttman scale is the
Coefficient of Reproducibility, which is:
Coefficient of Reproducibility = 1 - number of errors / number o f responses,
W here the number of errors refers to the number of observations that do not
conform to the order of responses being tested. A score of .9 or higher indicates a
113
unidimensional Guttman-type scale (Miller, 1991). Tables V-7 (Have Difficulty),
V-8 (Unable), and V-9 (Get Help) present coefficients of reproducibility for two
representative orderings for the IADLs over four years for each question format.
These tables suggest, echoing Lazaridis’ analysis of ADLs, that alternate orderings
do no better or worse.
The conclusion to be drawn from this exercise and the results reported by
Lazaridis et al. (1994) is that there is not a single unique pattern that characterizes
loss of ADL or IADL task ability. Using these items as a Guttman scale would
only be appropriate for a very general purpose. Disability in one task does not
imply disability in any other task. Thus, if one chooses to use these items in a
Guttman scale, one will essentially be glossing over the diversity in patterns of
change in functioning among the older population. This would result in an
important loss of information. Thus, in response to the second question, ADLs
and IADLs are not Guttman hierarchies, and in most cases should not be used as
such. This raises the interesting question, however, of how abilities are lost. For
example, do certain diseases have progressions of ability loss associated with them?
That is, for populations with a specific disease profile would loss of abilities occur
in a predictable order? Or, might groups of tasks be gained and lost in clusters?
If so, what characterizes the clusters and what other health characteristics are
associated with their loss?
114
The Relationship Between the ADL Scale and the IADL Scale
The factor analyses indicate that the items from the ADL scale reflect one
dimension of functioning, and items from the IADL scale reflect another dimension
of functioning. But how are those two dimensions related? In practice, ADLs and
IADLs are generally assumed to have a hierarchical relationship — that is, people
are assumed to lose their ability to do IADLs first and then their ability to do
ADLs. To investigate this assumption, proportions of persons experiencing decline
and improvement in each function over each interval are examined, and then a
cross-tabulation of disability in ADL by disability in IADL are presented. If the
two scales do measure different levels of functioning, with IADLs representing a
more strenuous or difficult level of tasks than ADLs, then we would expect to see
(a) that people are more likely to lose ability in the more difficult tasks first, and
(b) that people without disability in IADL would also be without disability in ADL
tasks, and that people with ADL disability would have disability in IADL as well.
Table V-10 presents the proportion of person reporting loss of ability to
perform each of the ADL and IADL activities over each of the three intervals.
People are more likely to lose IADL functions than ADL functions, suggesting that
IADL tasks are more difficult than ADL tasks. An exception to this association is
inability to use the phone; while using the phone seems to fit in conceptually with
the other IADL tasks, the frequency with which people become unable to use the
phone indicates a level suggests that it is more akin to the ADL tasks. With a few
exceptions, mostly between adjacently ranked activities, this ordering is constant
115
across the intervals. In Table V - ll, the proportion of persons regaining ability to
perform each item are listed for each interval in descending order. The order of
the items is the same as for loss of functioning, reflecting the importance of the
distribution on propensity for subsequent change — those who have not previously
lost functioning cannot regain the ability. Similarly, these distributions suggest
that, propensity to regain function being equal across activity items, those activities
that more people are unable to perform will be the same functions that more people
regain (because a function must be lost before it is regained).
The cross-tabulations between the number of ADL impairments and the
number o f IADL impairments in Table V-12 shows more detail about the
relationship of ADLs to IADLs. Looking at the first column, as the number of
ADLs increases, the chance that a person will have no impairment in IADLs drops
off precipitously, such that once a person is unable to do three or more ADLs it is
most unlikely that they will not have some IADL impairments. The data in the
second column indicates that a person with no IADL impairments almost certainly
will not suffer from ADL impairment; one can accumulate IADL impairments
without ADL impairments. But, as the number of IADL impairments increases,
the chance that a person will have no ADL impairments declines.
These data indicate that ADLs and IADLs are ordinally associated, but that
there is not a clear pattern. That is to say, a person does not lose all IADLs then
go on to lose ADLs. Rather, this table, and the results from the Guttman exercise,
indicate that there are multiple processes of change in functioning that are
116
characterized by loses of selected combinations of items, perhaps progressively,
from the ADL and IADL scales.
C. Implication o f Question Wording
Within the category of self-reported functioning measures, and specifically
among ADL and IADL items, there are several variations on how questions are
asked in national surveys of the older population. Some of these variations suggest
a flawed questioning strategy; other variations represent legitimate choices in how
a question is asked. A flawed strategy, for example, is apparent in questions that
do not qualify the typical "do you have difficulty w ith..." component to indicate
"because of health" or "some other reason." Similarly, the "how much
difficulty..." component sometimes neglects to clarify whether an "unable"
response indicates whether the respondent is supposed to answer "unable to do this
activity without any mechanical assistance (or) without help," or whether the
question is asking for ability with any usual personal or mechanical assistance.
Survey items that ask questions such as these present serious difficulties for
interpreting responses.
On the other hand, the specific questions may vary legitimately, sometimes
querying presence of difficulty, level of difficulty, or need for or receipt of help.
These queries, while asking for similar information, are indeed different, and will
lead to different responses from persons with the same level o f functioning. Also,
some surveys choose to collect information about ability without mechanical or
personal assistance (e.g. The Longitudinal Study on Aging) while others focus on
117
disability even when personal and mechanical assistance are used (e.g. The Survey
of Asset and Health Dynamics of the Oldest Old, Wave I). The choice of question
type depends in part on the substantive issue the researcher is interested in, and for
those doing secondary analysis, on data availability; but the choice also has
implications for the comparability of the study. Although these differences in
question wording do not necessarily introduce a reliability problem, when the
measures are used comparatively it does lead to variations in estimates of
prevalence of disability (Guralnik & Simonsick, 1993; Rodgers & Miller, 1994).
Therefore, since three versions of the question are available in this study,
differences in the rates of change in functional status for these question types can
be evaluated. First, issues and previous research on response bias by sex and race
are reviewed. Then, differences in the rates of change for the three question
formats are compared.
1. Response Bias
When the question on difficulty or receipt of assistance with a specific task
has not clearly included a "because of health reasons" clause, systematic
over/under-reporting by gender because of traditional sex role divisions of labor
occurs. Over-reporting occurs when men report receiving assistance with such
tasks as preparing meals that they have never done themselves regardless of health.
They may also underreport having difficulty with the task because they have had
no occasion to attempt it recently. Allen, Mor, Raveis, & Houts (1993) find that
men (80%) are much more likely than women (30%) to attribute help received to
118
such a longstanding allocation of chores. And, while this relationship does not
vary by age or education, the differences become trivial for gender neutral tasks.
Thus, if and when gender role division of household labor breaks down, this
relationship may change among future cohorts.
Beyond gender role division of labor, however, there may be other gender-
related bias in reporting, with or without appropriate qualifications to the
questions, and there may also be legitimate differences in reporting disability if the
process of loss of ability occurs differently in men and women. Evidence suggests
that decrement in functioning may occur differently in men and women. Some of
the confusion about sex and functioning ability has arisen because men have higher
mortality rates than women, but women report worse functioning. This has
appeared contradictory to some people, and thus has suggested response bias. On
closer inspection, however, the relationship is explainable. Crimmins and Saito
(1993) find that women are more likely than men to report loss in Nagi-type items
(doing heavy housework; light housework; walking a quarter of a mile; walking up
10 steps without rest; reaching out as if to shake hands; reaching up over the head;
stooping, crouching, kneeling; lifting 10 pounds), but that men are more likely to
report difficulty with ADL and IADL items. But, these declines among men may
be shortly followed by death leaving a higher prevalence of functioning impairment
(but as the work by Crimmins and Saito indicates, at a lower level of disability)
among the surviving women (Manton, 1988). And, when the questions include the
"because of health" clause, sex related response bias appears negligible.
119
Impairment in ADL and IADL are strong predictors of mortality, equally so for
men and women (Manton, 1988). While this evidence suggests that there is
minimal, if any, gender bias in reporting on these items, there is no good strategy
to settle this question conclusively.
Another potential source of systematic reporting bias is cultural. Although
a number of studies have found differences in health reports by race (for example,
Berkman, Singer, & Manton, 1989; Ferraro, 1987, 1993), few have attempted to
systematically determine the source of these differences: whether there are distinct
health differences or whether there are differences in reporting behavior by
race/ethnicity. Gibson (1991) is among those who have attempted to directly
examine the source of the differences in reports of disability by race. She
examined race differences in reports of various health measures in a measurement
model using LISREL. Although the results of her analysis on ADL and IADL
items are somewhat ambiguous, they do indicate that there is not a difference in
how underlying health status influences the response to ADL and IADL questions.
Nevertheless, Gibson (1991) did find evidence that there is some difference in what
influences responses to ADL and IADL questions from African Americans and
whites. Gibson (1991) suggests that the unmeasured influence causing the
difference in disability reports by race is probably not related to race itself, but to
some correlated characteristic(s). She hypothesizes that these characteristics
include social and economic characteristics such as work status, income, poverty,
and education.
120
2. Evaluation of the Impact o f Question Type
on the Representation of Change in Functioning
In this section, the logical and empirical relationship between having
difficulty, receiving help, and being unable to do alone the ADL and IADL tasks is
evaluated. This analysis provides relational information on response to the three
question formats, which is useful both for choosing a measure and for gauging
comparability with other studies. If we are only interested in receipt of help for
activities a person would have difficulty performing alone (i.e. for health reasons),
we would expect prevalence of difficulty to be at least as high, but probably
higher, than prevalence of receipt of help. Because the receipt of help questions in
the LSOA are only asked when a person has already indicated that they have
difficulty with an activity, this relationship will exist by construction. Degree of
difficulty (some, a little or a lot) is also only asked of persons who report having
difficulty with an activity. Thus, the number of persons reporting inability to
perform a task may be as high, but not higher than the number reporting difficulty
with the task. The relationship between inability to perform a task and receipt of
help is not determined in the question sequence, and although a relationship may
be assumed to exist for certain "essential" activities such as toileting, eating and
transferring, in reality the congruence between disability and receipt of help for
these items is imperfect. And, although a general ordering can be identified
logically in this way, the exact magnitude of difference cannot. Therefore, in this
section the relative differences in transition rates between having no difficulty and
121
having difficulty; being able and being unable; and not receiving help and
receiving help are compared.
Sample
This analysis uses the same sample that has been used in the previous
investigations (n =2,430). However, because the transition from functioning status
at one time to functioning status at the next time is of interest, the data have been
pooled to create 7,029 person-observations. Manipulating the data structure in this
way maximizes the number of observable transitions.
Measures
Each record now contains the person’s age and the indicator for that
individual’s functioning status at the beginning of the interval and at the end of the
interval. Age is adjusted to reflect age at the middle of the interval, when the
transition is assumed to occur.
The goal of this analysis is to assess how the different ways of asking the
question of disability in ADL and IADL produce different results. Specifically, it
is hypothesized that persons do not report having difficulty at the same rate at
which they report being unable; nor are people expected to report being unable at
the same rate at which they report getting help, and this analysis will indicate how
much these question formats differ from each other. Measures based on each
of the three question variations available in the LSOA are constructed. These
measures indicate whether or not the respondent (1) has difficulty with any ADL,
122
(2) is unable to perform any ADL, (3) or receives help with any ADL. Another
three such measures are also created referring to ability in IADL.
Analysis
To estimate transitions between having no limitation and having limitation
for both ADLs and IADLs, for each question format, a continuous-time hazards
model is estimated using SAS PROC LIFEREG (SAS Institute Inc., 1989).
Although this method assumes continuous time observation of a continuous process
and the data afford only fragmentary observation, the assumption of the analysis is
that the rates of transition are constant over each interval of observation, and that
the transitions that occur happen at the midpoint of each interval. Thus, with
discrete observations it is possible to approximate a continuous process with this
method.
Results
Figure V -l presents plots of transition rates from a hazard analyses for loss
of IADL function for each of three different question types: having difficulty,
being unable, and getting help. The plot lines indicate age-specific rates at which
a person becomes unable, begins to get help, or begins having difficulty with
IADLs. There is very little difference between having difficulty and getting help;
persons become unable to do IADLs at a somewhat lower rate.
For ADLs, however, the differences are more pronounced (Figure V-2):
The rate of incidence of having difficulty with an ADL is much higher than the
rates at which one becomes unable or begins to get help; however, the rate of
123
incidence of getting help is also considerably higher than that for becoming unable.
Further, the differences in risk of each type of loss increase as people age. These
figures show that different questions will yield different rates of incidence of loss
o f functioning. Because studies variously use each of these question formats to
represent functioning ability, comparability of results must be done with caution.
And, although in the LSOA there is a structured association between receipt of
help and having difficulty, and being unable and having difficulty (only those who
have difficulty are eligible to report receipt of help or disability), the underlying
causal mechanisms determining whether or not a person reports disability or
receipt of help are suspected to be different. Receipt of help is dependent on a
number of factors independent of health, including availability of help-givers and
financial resources. Further, when receipt of help measures are used exclusively,
those who receive help are obviously the object of the analysis. This is an
imperfect representation of those who need assistance. Therefore, because the
focus of this study is on the causes of maintenance and change in functioning,
difficulty and disability are more appropriate. Unlike receipt of help, indicators of
difficulty and disability can be used to infer who requires support.
D. Discussion
A review of the literature on the use of ADL and IADL scales reveals that
there has been considerable research on the properties of the items from the ADL
and IADL scales as measures of functioning status, but very little investigation into
how they perform as measures of change. And, the research on the properties of
124
ADL and IADL scales used to indicate status leaves unresolved questions regarding
the behavior of the individual items comprising these scales and the general use of
the scales. The issues addressed include the dimensionality of the ADL and IADL
scales; and the relationship of the items within each scale to each other
(scalability); the relationship of the items on one scale to the items on the other;
and the impact of differences in question format on the representation of change.
Each of these issues is investigated in analytic sections of this chapter.
The results from the analyses presented in this chapter suggest that ADL
and IADL are two distinct dimensions, and that while the items within each scale
are not hierarchically associated, that the two scales themselves are related in a
loose hierarchy. While these properties hold generally for each of the question
formats, there is apparently less "noise" in the "receive help" and "unable"
versions o f the questions. However, receipt of help and disability are theoretically
different constructs, And, a comparison of incidence of receipt of help and
disability indicates that there is an empirical discrepancy as well. Persons are
somewhat more likely to receive help than to become unable to perform a task.
This difference is more pronounced for ADL tasks than IADL tasks. The reason
for this may be the relative ease of obtaining assistance with IADLs compared to
ADLs. Details of these findings and the implications for the construction of the
dependent variable in this study are described below.
The analyses presented in this section lead to several conclusions regarding
how items from ADL and IADL scales should be aggregated to measure change in
125
functioning, and how differences in question wording affect the representation of
change in ability. First, dimensionality of functioning measures is considered.
This analysis established that ADL and IADL functions form two distinct
dimensions of functioning. These dimensions remain stable over time as a sample
population ages; the dimensionality is consistent regardless of whether have
difficulty, disability, or receipt of help measures are used. This analysis
establishes that scores indicating change in ADL and IADL ability also lie along
two dimensions, and similarly, that these dimensions remain stable over time as the
sample ages. Nevertheless, the structure is more clearly defined in both the
analysis of status and change for the disability and receipt of help questions
(compared to the have difficulty questions).
The next question is whether within each of the two groups of functions,
ADL and IADL, it is reasonable to consider the items to be hierarchically related.
Building on the work of Lazaridis and colleagues, this analysis suggests that, like
ADL items, IADL items ought not to be used in a Guttman-type scale. A variety
of orderings work equally well, indicating that there is no single process of
cumulation of functional difficulty. It would be incorrect to infer that an individual
who is unable to do a specific ADL or IADL task is also unable to do another task
on that battery. This finding raises the new question of how these task functions
are lost and how disability accumulates. There is clearly considerable diversity in
the processes of functioning change at older ages.
126
Approaches to combining the ADL and IADL items together treat the two
scales as hierarchically associated. This analysis finds that in fact there is an
ordered association between the two scales. Importantly, however, this is not a
strict association: The process of disablement is not such that an individual loses all
IADLs and then goes on to lose ADLs. Rather, here again, there is evidence that
there are a variety o f processes of functioning change that each involve selected
ADLs and IADLs, but only in certain cases are all IADLs or all ADLs lost. This
loose ordering suggests that the IADL tasks are more strenuous or involve different
(and more difficult) functions than ADLs. That there is evidence that IADLs are
lost first or more frequently suggests that IADLs have more taxing requirements,
but the variability in this association suggest that in some ways ADLs and IADLs
may represent different kinds of abilities.
This much establishes the structure of the items and their relationship to
each other. It also compares the several approaches to measuring functioning
ability by repeating the analysis of dimensionality and scalability for each of the
three question types. Although the three approaches to measuring change (having
a disability, having difficulty or receiving help) have similar structure, the rates of
incidence of disability, difficulty, or receipt of help are quite different, especially
for ADLs. This means that studies using the different measures may end up with
quite different conclusions about rates of functioning impairment.
The items on receipt of help, although clearly related to the more direct
indicators of functioning impairment, ask a substantively different question. It is •
127
expected that the correlates and predictors of receipt of help would include a
variety of other types of characteristics in addition to those shared with the
indicators of physical ability. Because the focus of the present study is on changes
in physical functioning, the use of the receipt of help questions would be
inappropriate.
For each of the question formats more people report no disability than
disability. The skew toward no disability is larger when the "receipt of help" or
"unable" formats are used; the indicator of whether or not an individual has
difficulty with the ADL and IADL items has a more balanced distribution. The
advantage o f having a more balanced distribution is that more changes to disability
would presumably be observed, and this would result in a more robust analysis.
However, recent studies that purport to find evidence of unreliability in
ADL and IADL items (Mathiowetz & Lair, 1994; Rodgers & Miller, 1994) rely on
responses to "have difficulty" questions. The face validity of the different ways of
asking the questions would suggest that disability (unable to do) items would be
considerably less vulnerable to such bias. Indeed, in the series of factor analyses
presented for both the status items and the change indicators, the loadings of the
ADL and IADL difficulty items was much less neat than the loadings for the
disability items. This suggests that the difficulty questions may not be capturing
the ADL and IADL constructs as clearly as the "unable" questions. This may be
because there is more measurement error in the difficulty items. The concept of
having difficulty is more ambiguous, and the difficulty questions may elicit a more
128
subjective, and possibly less reliable, response than the questions on being unable
and receiving help. For these reasons, responses to the unable questions will be
used as indicators of disability in the remainder of the study.
Although analyses on the 12 ADL and IADL items clearly rule out any sort
o f ordered association within the ADL and IADL scales, an ordered association
between ADL and IADL functions is supportable. Therefore, to make a coarse
representation of a progression of disability, a three-level response variable
indicating (1) not unable to perform any ADL or IADL item, (2) unable to perform
at least one IADL item, but not unable to perform an ADL item, and (3) unable to
perform at least one ADL item will be adopted.
The analysis presented here, in addition to informing the decision on how to
aggregate the ADL and IADL items also has important implications for the
interpretation of the summary measure. The factor analysis clearly indicates that
the ADL and IADL items form two dimensions that coincide with the ADL and
IADL scales. This suggests that ADL and IADL items represent categorically
different aspects of functioning. Thus, ADL and IADL may be seen as two
qualitatively different types of dependency.
Alternately, however, the ADL and IADL scales were shown to be
hierarchically related. This ordinal association is both empirical (people who have
ADLs also have IADLs, but those with few IADLs are unlikely to have ADLs) and
also substantive: ADLs, indicating ability for personal care, represent a more basic
level of disability than IADLs, indicators of ability to maintain a household.
129
Therefore, while recognizing the substantive implications o f the categorical
differences between ADL and IADL disability, in this research the two levels of
disability will be treated as ordinally associated.
130
Transition R a te s
Figure V—1: Transition Rates
Loss of IADL — Three M eosures
0.4
0 .35
0.3
0 .25
0.2
0 .15
0 .05
0
7 0 7 2 7 4 7 6 7 8 80 82
□ Difficulty
64 86 8 8 90
Age
Unable
9 2 94
Get* Help
131
Transition R a te s
Figure V—2: Transition Rates
Loss of ADL — Three M easures
0 .4 5
0.4
0 .35
0.3
0 .2 5
0.2
0 .1 5
0.1
0 .05
0
7 0 7 2 74 7 6 78 8 0 82 86 94 9 6 98 84 9 0 92
□ Difficulty
Age
o Unable Get Help
132
Table V-l: Exploratory Factor Analysis with 12 Measures of Functioning Difficulty from ADL & IADL Scales
at_4=Intervie^_St|mda^_Re2re£sion_£o^ficients_^rjtolic^e_iR^ationj_^unded_and_^i^tiglied_b^^0^^_^_
1984 1986 1988 1990
Activity Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor
Bathing 64 51 58 49
Dressing 47 52 55 72
Transfer 70 72 69 60
Toilet 54 49 47 75
Eating 38 42 24 55
Walking 74 69 72 37
Cooking 62 65 71 76
Shopping 41 59 55 79
Phone 62 54 65 50
Mng Money 78 77 78 68
Lght Hswk 43 51 58 64
Hvy Hswk 58 49 52 63
OJ
W
Table V-2: Exploratory Factor Analysis with 12 Measures of Functioning Disability from ADL & IADL Scales
1984 1986 1988 1990
Activity Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor 2
Bathing
Dressing
Transfer
Toilet
Eating
Walking
58
82
86
82
65
53
43
77
85
71
59
41
46
82
87
41
59
42
72
84
85
67
55
Cooking 69 71 76 79
Shopping 76 80 79 83
Phone 26 38 37 39
Mng Money 71 63 69 65
Lght Hswk 65 66 61 71
Hvy Hswk 50 56 60 69
Table V-3: Exploratory Factor Analysis with 12 Measures of Receipt of Help with Activities from ADL & IADL
j^jalesnoat_4-Jhiterviews_St|yid!y£di jlecy£ession>Coefficijynts_for_0^
1984 1986 1988 1990
Activity Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor
Bathing 61 44 44 44
Dressing 74 67 70 69
Transfer 85 65 80 74
Toilet 86 79 86 87
Eating 53 59 48 65
Walking 38 32 36 34
Cooking 62 65 67 72
Shopping 71 77 81 78
Phone 56 37 37 36
Mng Money 68 68 63 56
Lght Hswk 58 53 56 62
Hvy Hswk 52 57 63 67
U i
Table V-4: Exploratory Factor Analysis with 12 Measures of Change in Functioning Difficulty from ADL & IADL
Scales_a^_3_Intervals_§tandard- Jle2£essioii_Coefficients_^or_0bli2uei J'lotationJ t - ^O2mded_and_Multi£lied_b3^1 () ( )
Interval 1___ ____Interval 2___ ____Interval 3____
Activity Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor 2
Bathing 55 60 62
Dressing 63 64 47
Transfer 57 48 59
Toilet 38 58 67
Eating 25 22 57
Walking 46 49 66
Cooking 59 63 64 52
Shopping 64 56
Phone 37 38 48
Mng Money 70 72 49
Lght Hswk 41 37 56
Hvy Hswk 29 38 44
Table V-5: Exploratory Factor Analysis with 12 Measures of Change in Functioning Ability from ADL & IADL
Scales at 3 Intervals Standard Regression Coefficients for Oblioue Rotation
Activity
Bathing
Dressing
Transfer
Toilet
Eating
Walking
Cooking
Shopping
Phone
Mng Money
Lght Hswk
Hvy Hswk
Interval 1 Interval 2 Interval 3
Factor 1 Factor 2 Factor 3
86
89
43
73
70
36
Factor 1
43
81
92
90
*
58
Factor 2 Factor 1
40
Factor 2
65
58
69
71
39
48
37
67
59
56
47
69
68
24
57
59
46
77
79
37
65
77
58
u>
~ o
Table V-6: Exploratory Factor Analysis with 12 Measures of Change in Receipt of Help from ADL & IADL Scales
a t _ 3 _ I n te i2 r iy ji_ _ S ta n d c y 2 lJ ^ c y £ e s s io i^ jC o e f f ic ie n ts _ jf o r _ O b li2 u e _ I to ta tio iy -M Roundec^-an<^> M u lti£ lie d _ j> 2 ^ ^ 0 0 > ^ _ _ - — > <
Interval 1 Interval 2 Interval 3
Activity Factor 1 Factor 2 Factor 1 Factor 2 Factor 1 Factor 2
Bathing
Dressing
Transfer
Toilet
Eating
Walking
50
74
58
85
54
34
52
77
75
85
19
47
53
86
68
91
39
45
Cooking
Shopping
Phone
Mng Money
Lght Hswk
Hvy Hswk
54
64
19
61
46
44
66
73
31
58
56
54
61
58
37
54
59
55
W
oo
Table V-7: Coefficients of Reproducibility1 for Alternate Orderings of
Have Difficulty Measures of Six IADL Functions at Four Interviews
Order:
Order:
Have Difficulty
Heavy Housework, Shopping, Cooking, Using
Phone, Managing Money, Light Housework
Year Reproducibility
1984 .93
1986 .84
1988 .81
1990 .78
Heavy Housework, Shopping, Using Phone, Cook,
Managing Money, Light Housework
Year Reproducibility
1984 .93
1986 .83
1988 .81
1990 .77
1 Table IV-7 - Table V-9: Coefficient of Reproducibility = 1 -
number of errors / number of responses. A score of .9 or higher
indicates unidimensional Guttman-type scale.
Table V-8: Coefficients of Reproducibility for Alternate Orderings of
Disabilit^> Jleasures_ofi _Six_i i IAMj_Function£_atu i - Four_Interview^^^_____
Order:
Order:
Unable to Perform Activity
Heavy Housework, Shopping, Using Phone,
Cooking, Managing Money, Light Housework
Year Reproducibility
1984 .96
1986 .94
1988 .90
1990 .88
Heavy Housework, Shopping, Managing Money,
Light Housework, Cooking, Using Phone
Year Reproducibility
1984 .97
1986 .95
1988 .92
1990 .89
140
Table V-9: Coefficients of Reproducibility for Alternate Orderings of
Recei£t_of__Hel£> Jleasures_£f_Si3c- i i - IMIj_FHinctions_j i at_>Four_J^terview^- __
Order:
Order:
Receive Help with Activity
Heavy Housework, Shopping, Cooking, Managing
Money, Light Housework, Using Phone
Year Reproducibility
1984 .95
1986 .89
1988 .87
1990 .83
Heavy Housework, Shopping, Light Housework,
Cooking, Managing Money, Using Phone
Year Reproducibility
1984 .95
1986 .89
1988 .87
1990 .83
141
Table V-10: Proportions Declining for Individual Functions Across Three
Intervals. Decline is Measured by Change Between Being Able and Being
U naiy;e < J to _ P e rfo rm _ J ^ h e _ A c tiv it£ i==iii— _ii====__ji=______= _=ii___i i - i _ _ _ _ i _ i=_ _ > i _ _
Activities ordered by percentage of sample experiencing loss of function
in the first interval; descending order.
Activity Scale 84=>86 86=>88 88=>90
Hvy Hswk IADL 13.1% 9.8% 12.8%
Shopping IADL 5.0% 5.5% 8.1%
Mng Mny IADL 3.0% 3.5% 5.0%
Lt Hswk IADL 2.3% 3.1% 4.9%
Cooking IADL 2.2% 3.1% 5.3%
Walking ADL 2.0% 2.3% 4.3%
Bathing ADL 1.9% 2.5% 4.9%
Phone IADL 0.3% 2.0% 3.6%
Toilet ADL 0.8% 1.0% 2.1%
Transfer ADL 0.8% 1.1% 2.1%
Dressing ADL 0.6% 1.1% 2.0%
Eating ADL 0.1% 0.2% 0.9%
142
Table V-ll: Proportions Improving for Individual Functions Across Three
Intervals. Improvement is Measured by Change Between Unable and Being
Able to Perform the Activity.
Activities ordered by percentage of sample experiencing recovery of
function in the first interval; descending order.
Activity Scale 84=>86 86=>88 88=>90
Hvy Hswk IADL 3.1% 6.9% 6.5%
Shopping IADL 1.5% 2.4% 3.0%
Mng Mny IADL 0.6% 1.6% 1.8%
Lt Hswk IADL 0.6% 1.0% 1.6%
Bathing ADL 0.6% 0.9% 1.5%
Walking ADL 0.6% 1.4% 1.2%
Cooking IADL 0.4% 0.9% 1.3%
Phone IADL 0.3% 0.7% 0.9%
Dressing ADL 0.2% 0.3% 0.5%
Toilet ADL 0.1% 0.4% 0.7%
Transfer ADL 0.1% 0.4% 0.6%
Eating ADL 0.1% 0.0% 0.1%
143
Table V-12; Relationship of ADLs to IADLs
Percentage of Those Having
no IADL Disabilities by
Number of ADL Disabilities
Percentage of Those having
no ADL Disabilities by
Number of IADL Disabilities
1984 1984
0 - ADL
1 - ADL
2 - ADL
3 - ADL
4 - ADL
5 - ADL
6 - ADL
89%
30%
2 0 %
0 - IADL
1 - IADL
2 - IADL
3 - IADL
4 - IADL
5 - IADL
6 - IADL
100%
92%
87%
84%
46%
71%
1986 1986
0 - ADL
1 - ADL
2 - ADL
3 - ADL
4 - ADL
5 - ADL
6 - ADL
80%
24%
6%
0 - IADL
1 - IADL
2 - IADL
3 - IADL
4 - IADL
5 - IADL
6 - IADL
99%
96%
85%
48%
45%
2 0 %
22%
1988 1988
0 - ADL
1 - ADL
2 - ADL
3 - ADL
4 - ADL
5 - ADL
6 - ADL
78%
6%
7%
0 - IADL
1 - IADL
2 - IADL
3 - IADL
4 - IADL
5 - IADL
6 - IADL
100%
93%
69%
69%
48%
21%
2 0 %
1990 1990
0 - ADL
1 - ADL
2 - ADL
3 - ADL
4 - ADL
5 - ADL
6 - ADL
74%
16%
6%
5%
19%
4%
0 - IADL
1 - IADL
2 - IADL
3 - IADL
4 - IADL
5 - IADL
6 - IADL
98%
92%
73%
49%
51%
15%
31%
144
VI. PATTERNS OF CHANGE IN FUNCTIONAL HEALTH
A. Introduction
The conceptualization of change in functioning (Chapter II) introduced the
notion that changes in functioning status are ongoing and create unique profiles of
individual health. That discussion also noted that the conventional approach to
representing change in status uses a transition analysis approach, restricting the
observation of change to two points. Here it is argued that understanding an
individual’s health status and chances of future change requires the observation of
history of change. Such data provide a different perspective on change in
functioning. In this chapter, an approach using individual histories of change
provides an extensive description of the experience of this sample of older
Americans. A typology of these histories is constructed and used in a multivariate
analysis that predicts types of changes in functioning based on the six-year history.
The study of health change in older persons has shown that while, in
general, physical functioning abilities decline with age in the population, this is
neither a universal nor an inevitable trend. The present analysis contributes to
previous work in three respects. First, in making use of four waves of longitudinal
data, it is possible to establish a sequence o f change between levels of functional
health status, whereas previous work has concentrated largely on transitions
between two states. Second, the analysis includes a preliminary exploration of
possible covariates of these sequences of changes. Previous work has rarely tested
the effects of the various predictors within a fully specified multivariate
145
framework. Here, a full complement of predictor variables is included in a
multivariate analysis. Third, functional health is measured in such a way as to
distinguish between levels of disability.
There has been a growing trend among researchers to disaggregate
populations and look behind simple prevalence rates in the cross-section to uncover
distinct patterns that are responsible for the heterogeneity of individual experience.
It has been the progress of such research that has lead to the recognition that the
association between older age and disability does not mean that as one grows older
one will become disabled and that the decline will be precipitous. Although this
pattern holds for the majority of the population, there are a substantial number who
experience improvement in functional abilities over time (Branch & Ku, 1989;
Crimmins & Saito, 1993; Rogers, Rogers, & Belanger, 1992). In the present
study, observations of functioning status on a group of elderly who survive from
1984 through 1990 are linked to trace individual sequences of change in level of
functioning. This approach allows differentiation, for example, between those who
do experience a decline in functioning from which they do not recover during this
period of observation, from those who decline and then regain functioning.
Building on previous research, a set of variables indicating social,
economic, and demographic status, the presence of disease, and other health state
indicators are proposed as correlates of different patterns of transition between
levels o f functional status. This research contributes to the understanding of
146
covariates of change in health with age by determining important differences in
these patterns.
B. Background
Research on change in physical functioning has focused almost exclusively
on transitions in status between two intervals. An important exception to this is
the recent work of Verbrugge, Reoma, and Gruber-Baldini (1994). In that study,
diary entries provide continuous data on changes for several dimensions of
functioning over a one-year period following hospital admission for a small sample
of persons 55 years of age and older. The results of this study indicate
considerable dynamism and heterogeneity in patterns of change in functioning
status over the year of observation.
Other research focusing on transitions between levels of functioning status
over longer intervals have found a large proportion of the change occurring for the
worse, with a smaller but important proportion of the sample reporting improved
functioning. Crimmins and Saito (1993) used the baseline and first wave of
follow-up interviews in the LSOA to consider the determinants of functional
improvement and deterioration in the ability to perform specific individual Activity
of Daily Living, Instrumental Activity of Daily Living, and Activity of the Wider
World tasks among older persons. This analysis revealed that there is a
considerable amount of gross change, both improvement and decline, in functional
abilities among older non-institutionalized persons. Branch and Ku (1989) and
Manton, Corder, and Stallard (1993) examine transitions between levels of
aggregate measures of ADL and IADL functioning items. Although the
proportions making improvements and declines in these studies vary because of
differences in level of aggregation of the measure of functioning, and because of
differences in the interval length, the sense is similar: Among older persons,
decline in functioning is neither universal nor precipitous; there is considerable
individual improvement in the general population, despite a separate, dominant
trend toward worse health. (For a review of the conceptual basis for the approach
taken in this chapter, please refer to Chapter II).
The present analysis builds on this previous work, using an index measure
o f functioning status rather than individual tasks, and linking changes in level of
functioning over multiple observation points. This study is complementary in
examining the covariates of net change in aggregate level of functioning among
older persons.
C. Sample
The data for this study, the LSOA and linked Medicare Part A files have
been described extensively in Chapter III. For this analysis, functioning among
those who survive over the entire six-year period is of interest, so only
observations from individuals with responses at each of the four interviews are
used (n=2,426). This includes those who were institutionalized at any interview,
but eliminates the sample members who died between 1984 and 1990 ( n = 1,684),
and respondents who were not interviewed at one or more waves (n=742). In
order to examine functioning over six years, the sample must be limited in this
148
way; Therefore, the results are representative of those who are non-institutionalized
and 70 years o f age or older in 1984, and who survive and respond through 1990.
This sample is slightly younger, has a higher percentage of women, fewer people
in poverty, and has better physical functioning in 1984 than the original LSOA
sample1 (Table V I-1)). This selection does not change the proportion of African
Americans in the sample.
The LSOA has linked all Medicare Part A (inpatient hospital) Beneficiary
records to the responses of the relevant sample individuals. These records on each
hospital episode are used to establish whether or not an individual has been
hospitalized with certain conditions during the six year period. These records also
provide the cost associated with hospital episodes occurring between 1984 and
1990.
D. Representing Change in Functional Health
To capture the pattern of an individual’s history of change in functioning
ability, a categorical dependent variable is created that represents four possible
patterns of change. In previous studies, change in functioning status has been
explored through transition rates between states identified to represent salient
features of dependency (Crimmins, Hayward, & Saito, 1994). In this analysis,
1 Most of the differences between the two samples result from the differences in
the likelihood of death. While we know that nonresponse to the survey is higher
among those with impaired functioning, analysis has shown that correcting for
nonresponse influences neither the prevalence of disability (Chapter III), likelihood
that a functioning transition will occur, nor the mean level of functioning problem
(Mihelic & Crimmins, 1995).
149
similar states are used to create a variable that synopsizes individual experiences
over the four interviews and categorizes these experiences into four patterns of
change. A multinomial logistic regression analysis is then used to explore the
association of several sets of theoretically specified covariates with these patterns.
Each individual is assigned a score for functional health status at each
interview: 1984, 1986, 1988, 1990. These scores are based on the respondents’
self-report on whether or not they are unable to perform five Activities o f Daily
Living (bathing or showering, dressing, eating, getting in and out of bed or chair,
using the toilet) and five Instrumental Activities of Daily Living (preparing own
meals, shopping for personal items, managing own money, using the telephone,
doing light housework). Functioning status at each interview is categorized as: (N)
No difficulty with any of the activities; (I) Unable to perform one or more of the
Instrumental Activities of Daily Living (IADL); and (A) Unable to perform one or
more Activities of Daily Living (ADL). (A detailed discussion of the empirical
rationale for this measure is given in Chapter V).
1. Interpreting the Categories of Health Status
The advantage of this categorization of levels of functioning status is that it
distinguishes between two qualitatively different types of dependency. The first
type of dependency (ADL) includes those types of tasks that are most immediate to
personal care, and the dependency in which necessitates care that may differ in
intensity and frequency from those in the second group. The second category
(IADL), suggests the need for assistance that is not only less intensive, but also
150
more flexible. The relevance of such categorization is argued by Soldo (1985),
who found that characteristics of the type of tasks individuals require assistance
with are important determinants of the source of the care. In particular, the
differentiation between tasks that are "discrete" and can be scheduled, versus those
that require intensive care and are irregular and unschedulable, is relevant here.
Soldo found that dependency in care needs that are complex, inconsistent, or
require constant supervision, significantly increases the probability of use of long
term care.
At each interview respondents are characterized as having (1) No Disability,
(2) having IADL disability, and (3) having ADL disability. While the proportion
reporting no disability with any ADL or IADL drops from about 93% in 1984 to
73% in 1990, the small proportions reporting inability to perform IADLs (4.5%)
and ADLs (2.6%) in 1984 rise to well over 10 percent each by 1990 (Table VI-2).
Women are more likely to report disability than men, and the difference is greater
for ADL disability than for IADL disability. This difference increases in
significance as the sample ages.
2. Patterns of Change in Level of Functioning
By tracing an individual’s movements between these states at each of the
four interviews, patterns of change in level of functioning can be observed. Figure
VI-1 diagrams the possible moves that can occur between any two interviews. An
individual can move from any of the three levels of functioning ability to any other
151
level, or stay in the same place. In reality, of course, people can also die, but for
this particular analysis, those who become deceased have been excluded.
Mathematically, these transitions could produce 81 possible combinations of
functioning status over the four interviews. Table VI-3 displays the remarkable
diversity of patterns of change in functioning that are actually reported, a total of
66 among this sample representative of Americans who are 70 years of age and
older and noninstitutionalized in 1984, and who survive through the 1990
interview.
Characterizing Patterns According to Sequence of Transitions
The many patterns of health change presented in Table VI-3 were
categorized according to sequence of change: No change in status from interview
to interview; Improving, status has moved to a level o f less intense dependency in
activities, a trajectory involving at least one transition to improved functioning;
Declining, status has moved to a level of more intense dependency, a trajectory
involving at least one transition to worse functioning; Fluctuating, status does not
follow a "trajectory" pattern of either increasing or decreasing in intensity of
dependency, but fluctuates, both increasing and decreasing over the four interview
waves. Such a typology, while to some extent imposing a limited number of
variations on a very diverse experience, has the benefits of a) providing a
manageable and interpretable variable appropriate for use in multivariate analyses;
and b) representing important but nonmodal patterns of change that may otherwise
be obscured.
Overall, more than two-thirds of this older population reports no change in
function over the six-year period, and this is true for three-quarters of the men.
About one-fifth report declines (only 16% for men) and less than one percent have
trajectories that can be characterized as improvement. Roughly 10% report a
sequence of change in functioning that involves both improvements and declines
(Table VI-4).
For comparison, the proportion making transitions in level of functioning
between interviews are presented with the analogous patterns in Table VI-5. Since
it is not possible to "fluctuate" over an interval, the proportions are distributed
over three categories rather than the four pattern types. When the restriction of the
six-year trajectory is not imposed, the percent improving is at least double in each
interval; and the proportion declining is roughly half that of the six year sequence.
The proportion reporting no change is also considerably higher over the shorter
period.
The proportions for sequence of change differ from transition estimates by
restricting the definition of change to reflect a trajectory that is consistent over a
longer period. Also, this method of linking information over four waves yields
the added information on the proportion declining who then go on to improve, and
of those improving who decline again during the period of observation.
Patterns of Change and Functioning Status
The association between the sequence of transition and functioning health
status can be seen in cross-tabulations with the initial and final level of functioning
153
(Table VI-6). Among those who are 70 years and older in 1984 and who, at that
time, report being unable to perform one or more ADL, 50% will remain at that
level of impairment over the subsequent six years. Importantly, however, the
other half will experience some improvement in functioning; 19% will experience
improvement without declines over the duration of the observation, and 30% will
experience improvements followed by declines and possibly further improvements.
M ost of the elderly do not have any functioning impairments in 1984. Among
those who begin able to do all of the activities, over 70% will maintain that
superior level of functioning ability through 1990. Twenty-eight percent
experience some decline; about 8% declining and then improving again.
While those reporting no disability in 1990 are much more likely than those
who are unable to do any ADL or IADL activity to have had no change in status
over the four interviews, there is a clear pattern of association between worse final
status and having had a transition pattern characterized by deterioration in
functioning abilities.
E. Multivariate Analysis
In this section social, demographic, economic, health and health care
utilization characteristics are related to the pattern of change in functioning
observed in individual sample members categorized as improvement, decline,
fluctuation, and no change. The dependent variable has categories that are
assumed to be not correlated and not ordered. For this analysis a multinomial
(polytomous) logistic regression model will be estimated. For each type of change,
154
results will be presented in odds ratios, comparing the odds of having a given
pattern o f change (either improvement, decline, or fluctuation) to a comparison
category (maintaining a constant level o f functioning). This equation takes the
form:
In (P(Y =j) / P(Y =J)) = 0jo + /3 jIX 1 + ...+ /?jk Xk
W here Y is the outcome variable, J is the number of states in the outcome
variable, and j takes values from 1 to J -l. Ratios of probabilities relative to the
reference probability are estimated in the analyses.
The advantage of this type of analysis is that each category o f the dependent
variable may be regarded as qualitatively different, and it allows for differential
effects of the hypothesized covariates on each of these categories. Nested models
facilitate a comparison of the effects of sequentially introducing covariates and
observation of how these variables are related. Because significant differences in
the distribution over the types of change exist for these men and women who
survive for the six year period, the models will also be estimated separately by
sex.
1. Dependent Variable
As can be seen by the cross-tabulations between initial functioning status
and the sequence of change indicator in Table VI-6, there will be a population that
is not eligible to decline (those who begin with ADL disability), and another that is
not eligible to improve (those who begin with no difficulty in any of the activities).
Therefore, for the remainder of the analysis, when considering movement between
155
these types of categories, the sample will be selected to include only those who are
eligible to make the moves being considered. The dependent variable for the
decline analysis has three categories: decline, fluctuation, and no change; and the
dependent variable for improvement indicates improvement, fluctuation, and no
change2. The pattern of fluctuation is included in the analysis so that patterns of
improvement and decline may be clearly compared to patterns of no change.
Interpreting patterns of fluctuation is complicated by the change in sample between
the two analyses. Consequently, these results are not presented. However, results
from a parallel but separate analysis that identifies predictors of the probability of
fluctuation relative to having no change in functioning are presented.
2. Independent Variables
Predictors of change in functioning are chosen to reflect elements of each
component o f the conceptual model presented earlier (Chapter II). Measures
predicting pattern of functioning change come from two sources. The first are
LSOA respondent reports o f age, sex, race (African American and non-African
American), years of education, income (recoded to indicate whether or not the
individual is below the poverty threshold), whether or not the person lives alone,
whether or not the person has had arthritis in the past 12 months, has difficulty
2 Because there is such a small proportion of the sample improving according to
the definition used in this approach, the sample size precludes a reliable analysis, and
the results of the improvement analysis are not presented here. The approach used
in the next chapter does allow predictors of improvement to be compared to predictors
o f decline, however.
156
with vision or hearing, or walks a mile at least three times per week. Height and
weight are recoded to indicate percentile rank of the person’s body-mass ratio
(kg/m2 ). Functioning status at baseline is also controlled3. The conceptual model
of the process of disablement used here (Chapter II) distinguishes between
disability (inability to perform integrated tasks) from physical limitations (inability
to perform component operations that indicate less than ideal physical ability, but
not inability to carry out tasks required for independent living). Physical limitation
is included in this analysis using a measure that indicates how many of the
following items are difficult for the individual: walking a quarter of a mile;
walking up 10 steps without rest; standing or being on feet for about two hours;
sitting for about two hours; lifting or carrying 10 pounds; stooping, crouching or
kneeling.
Because there is some concern about the consistency of proxy responses
compared to self responses on level of functioning, a measure indicating whether
or not the respondent changed over the four interviews is also included. It is
possible that the use of proxy responses introduces measurement error, and some
research has found that caregivers (the most likely proxy respondents) have a
tendency to understate functioning ability (Magaziner, 1992; Rubenstein, Schairer,
Wieland, & Kane, 1984). Therefore, the indicator for having changed respondents
3 In both the analysis of decline and fluctuation indicators for beginning with
IADL impairment are included. In the analysis of fluctuation there is no "ineligible"
category, so a dummy indicator for beginning with ADL impairment is also included.
In both analyses the omitted category is having no disability.
157
is included to control for any such measurement error4. One third (34.3%) of the
sample members had a change in respondent. Because of the link between the use
of proxy respondents and poor health, a decline in functioning would be expected
with a change in respondent, regardless of measurement error.
Measures of diseases and impairments of the respondents are available from
two sources. At the outset of the six year period, respondents report whether they
have ever had, or have had in the past year, certain diseases: heart disease,
cerebrovascular disease, cancer, diabetes, osteoporosis, arthritis, and whether or
not they have difficulty with their vision or hearing. Over half (55.8%) report that
they have arthritis. A much smaller proportion have diabetes (7.7%), osteoporosis
(4.0% ), or difficulty with hearing (4.7%) or vision (6.2%) after correction or the
use of aids. Among conditions respondents report ever having had, heart disease
(14.2%) and cancer (10.8%) are most common, while only 4.3% report having
had a stroke or cerebrovascular accident.
The second set of measures are from Medicare Part A (Inpatient Hospital)
records. This provides an identifiable time and threshold for health events over the
period when functioning change is examined (10/1984— 9/1990). From these
records, diagnoses are extracted5 indicating inpatient hospital episodes associated
4 The measure used indicates whether the respondent changed between proxy and
self or between type of proxy (living in household or not).
5 Medicare Part A records are available for the duration of the LSOA survey,
1984-1990. For each Medicare Part A covered episode during this time, up to 5
diagnoses may be recorded. Diagnoses for this analysis are defined by the following
158
with heart disease, malignant neoplasm, atherosclerosis, cardiovascular disease,
and chronic obstructive pulmonary disease. This set of disease states is chosen in
part because of previous research suggesting an association between these
pathologies and functioning, because of their ranking among the most probable
causes of death among person over 70 years of age, and because there were
enough events to include each disease in the analysis. While there is a history of
differentiating fatal from disabling conditions, much remains to be understood
about the relation between life-threatening diseases, interventions, and subsequent
disability. Interestingly, there is almost no correlational relationship between the
prior presence of a disease and subsequent hospitalization for that disease.
From these files an indicator is constructed for the total cost of Medicare
covered episodes. This provides an indicator of the seriousness of the totality of
health problems that have been experienced by these people over the six years. On
average, the Medicare cost for hospitalizations over the duration of the survey
(10/84-9/90) was $9,160. For those who are covered by Medicare but did not
have any hospital episodes, the absence of Medicare spending is recorded.
Because not all persons were willing or able to provide the requisite information
for linking their Medicare records (n=57), and for some a match was not made
even though information was supplied (n=212), an indicator for those non-matched
ICD-9 codes for the primary diagnosis recorded during any Medicare Part A covered
episode: heart disease (391-398, 401-405, 410-417, 420-429); malignant neoplasm
(140-208); atherosclerosis (440); cardiovascular disease (430-437); and chronic
obstructive pulmonary disease (490-496).
159
persons is also included. Persons for whom no match was made are assumed not
to have had any of the episodes of interest and to have spent no Medicare dollars.
The indicator variable controls for errors in this assumption, as well as indicating
differences between persons for whom a match was not made and those whose
records were successfully accessed. Because including an indicator for intensity of
service utilization is relatively new to this type of analysis, this measure is added
in alone (with the control for non-matched cases) in a second model in each
analysis.
3. Analysis of Declines
The effect of the covariates on the risk of experiencing decline in
functioning relative to reporting no change in functioning is indicated by the odds
ratios in the first two columns of Table VI-7. In the first model for decline,
increasing years of age raises the chance that one will have a decline compared to
no change in functioning. Each additional year increases the chance by 12%.
Being female doubles the relative risk of experiencing decline relative to staying
the same. Additional years of education, being African American, and being in
poverty do not affect the relative likelihood of experiencing decline in the full
sample.
Those who had a heart disease related hospital episode are over 80% more
likely to have a decline in functioning relative to remaining at the same level of
functioning over the six-year period. Having been hospitalized for a
cerebrovascular condition more than doubles the relative risk of decline, while
160
hospitalization for chronic obstructive pulmonary disease increases it seven times.
Having a history of diabetes at the inception of the survey increases those odds by
about two times. Having a history of heart disease prior to the survey is associated
with a reduced chance of decline, with all these other characteristics taken into
account. This effect may be a result of having excluded persons who died before
the final interview in 1990, and because of beneficial effects of lifestyle changes
made in response to initial evidence of heart disease among the survivors (changes
such as improved diet, increased exercise, and so on). Vision impairment presents
a clear risk for functioning decline. Those who report having difficulty with their
vision are over six times as likely to report a decline compared to staying at the
same level of functioning. Being severely underweight increases the relative
chance of decline by 80%. An indicator of the number of physical limitations an
individual has difficulty with increases the chance of decline by about one third. A
change of respondent is associated with an increase of over eight times in the
chance of a pattern of decline relative to remaining at the same level. This is not
surprising given that the most common reason for changing respondents is the
decreased ability of the initial respondent to answer because of physical or mental
decline.
An indicator for intensity of service utilization or seriousness of health
problems is added in the second model predicting the relative risk of decline.
When dollars spent on hospital episodes is controlled, hospitalization for heart
disease is no longer a significant predictor of decline in functioning. All other
161
effects are stable. Those with large hospital bills are more likely to decline in
functioning. Those for whom matches with Medicare records were not possible
are also more likely to have patterns of declining functioning. It is a matter of
speculation who the no-match people are; That very few of the non-matched
people actually refused to provide information suggests that the problem providing
matchable information may be associated with ineffectiveness in access.
In Table VI-8, this same analysis is presented for subsamples by sex.
These analyses reveal differences by gender in the predictors of change in physical
functioning. The predictors of decline relative to no change for women are quite
similar to those for the full sample. Several differences exist, however. Very low
weight does not predict decline among women, nor does the control for having a
missing Medicare record match. And, although having had a hospital episode
associated with cerebrovascular disease is significant in the first model, when
hospital associated expenditures are controlled, this is no longer important.
For men, several interesting differences exist. First, number of years of
education is salient for men, but it is not for women nor for the full sample. For
men, each additional year of education reduces the chance of decline in functioning
relative to staying the same. Among men, the apparently compensatory effects of
baseline heart disease do not exist. And, diabetes is not significantly associated
with an increased chance of decline in functioning for men, as it is in women.
However, for men (but not for women) inability to provide adequate information to
162
match Medicare records is associated with a significant and fairly large increase in
the relative chances of decline.
4. Analysis of Fluctuations
Ratios indicating odds of fluctuation relative to having no change are
presented in the second two columns in Table VI-7. Persons who are older, have
had chronic obstructive pulmonary disease requiring a hospital stay, have difficulty
with their vision, a greater number of impairments, who changed respondents, and
who had greater hospital care utilization are more likely to have "ups and downs"
in functioning ability compared to staying at the same level. The effects of these
characteristics in predicting fluctuation compared to staying the same are similar to
their effects predicting the relative risk of decline. Exercise, however, appears in
this model as a predictor of maintenance of functioning relative to fluctuation.
Those who walk a mile at least three time a week are about half as likely to report
a bidirectional change in their functioning as they are to report no change in
functioning. IADL but not ADL impairment increases the relative risk of decline,
and, all other things equal, each additional impairment increases the chance of a
fluctuating trajectory by about 40%.
The analysis of fluctuation by sex again reveals several differences (Table
VI-9). Having a hospital stay associated with chronic obstructive pulmonary
disease is not important for women, but for men it increases the chance of
fluctuations in ability by over 20 times. And, it is women who reap the beneficial
163
effects of exercise, not men. Interestingly, IADL impairment increases the chance
of fluctuation among women, but ADL impairment increases this chance for men.
5. Factors Affecting Probability of Decline and Fluctuation
To examine how variation in significant predictors affects the probability of
having a particular pattern of change in functioning, the mean probability of having
each type of transition for selected categories of the significant independent
variables is estimated. In contrast to the odds ratios that indicate the relative effect
on a type of pattern of change, these probabilities are absolute. These probabilities
are calculated for the entire sample using the equation presented in Table VI-7,
with the control for sex.
The first column of Table VI-10 shows the mean probability of decline
given specific characteristics. For example, if the entire sample were 70 years old,
but otherwise had the observed distribution of characteristics, their probability of
having a pattern of decline over the six-year period would be 12%. If the sample
were 20 years older, that probability would increase to 38%. With these other
characteristics taken into account, women are about 40% more likely than men to
have a decline in functioning. Having a hospital episode associated with
cerebrovascular disease, having a history of diabetes, or being seriously
underweight at the initial interview are each associated with an increase of about
40% in the probability of decline compared to those without these conditions.
Those who required hospitalization for chronic obstructive pulmonary disease or
who reported difficulty with their vision at the initial interview in 1984 are twice
164
as likely to report a pattern of decline in functioning over the six-year period.
Each additional impairment is associated with a modest increase in the probability
of decline.
Most people who have one impairment have several impairments, however.
Having three impairments increases the chance of decline by about 50% compared
to those with no impairments; those who have difficulty with all six activities are
about twice as likely to decline over the six years. Similarly, incremental
increases in dollars spent on Medicare covered hospital stays are associated with
relatively small increases in the probability of decline, although those with very
large expenditures (i.e. $100,000) are over twice as likely to have declines as those
without any such expenditures.
Alternately, 20 additional years of age increases the chance of fluctuation
by about 60%; having had a hospital episode for chronic obstructive pulmonary
disease has about the same effect. Difficulty with vision increases the chance of
fluctuation almost 80% from 8% to 14%. While the effect of additional
impairments is large, the rate at which they increase the probability of fluctuation
is not quite as steep as it is for decline. And, although intensive health care
utilization is associated with greater chances of fluctuation, relatively small
increments in expenditure do not have a large effect.
F. Discussion
This research contributes to the understanding of covariates of change in
health with age by determining important differences in patterns of transition
165
between levels of functioning ability at older ages. This approach preserves the
individual history of functioning change over a six-year observation period. Ten
common measures of functioning ability are used to form four levels o f functioning
that, although ordinally associated, reflect categorically different groups in terms of
type o f disability and type of assistance required. Histories of changes between
these levels are then categorized according to the type of trajectory they reflect:
improvement in functioning; decline in functioning; a pattern o f fluctuation,
including both improvements and declines; and a history of no change in
functioning over the four interviews.
This approach produces 66 unique sequences of functioning status. When
these sequences are categorized, it is clear that the modal trend in functioning is
not decline, but rather stasis. The relatively low rates of improvement, and also
decline, in both transitions and patterns, is a result of using a comparatively high
threshold for change. Most frequently, studies of changes in functioning use
indicators of having difficulty with ADL and IADL tasks. In this study only
disability (being unable) was used, and this produces lower rates of both
improvements and declines. Incidence of difficulty is certainly greater than the
incidence of disability, but there may also be some random variation in response to
"difficulty" questions (discussed in the previous chapter) that contributes to the
higher rates of improvement and declines with those measures.
The distribution of the sample between the categories of improvement,
decline, fluctuation, and no change differs considerably from the distribution over
166
the three categories produced by a transition approach at each interval. The
percentages making transitions in each interval are of a comparable magnitude to
distributions reported using other samples of approximately the same age and with
intervals of differing lengths, taking into account the difference in measures. This
is an important distinction because it suggests something about the dynamics of
functioning change: While there may be a considerable amount of gross change in
any interval, the proportion making each type of change is quite different when
that change is recorded over a series of observations.
The effects of demographic, social and economic characteristics, presence
o f chronic conditions, hospital episodes associated with chronic diseases, intensity
of medical utilization, and indicators of health behaviors and level of disability and
impairment on the pattern of change recorded for sample members are examined in
a multivariate analysis. Each of these independent variables has been found to
predict level of functioning in the cross-section, and many of them are known to be
associated with changes in functioning over an interval. In this analysis, their
effects on patterns of change, with initial level of functioning controlled, are
examined.
Among social, demographic and economic characteristics, only age and sex
predict declines in functioning. However, when analyzed separately, education is
an important predictor of declines among men. This finding is interesting in
contrast with earlier results by Mor et al. (1989) which suggest that education was
167
only salient among women. The difference may occur because o f the longer
horizon on which change is observed.
As expected, several of the chronic conditions (diabetes and vision
impairment) are associated with increased chances of decline. Osteoporosis,
arthritis and hearing impairment are not, however. The lack of effect of those
chronic conditions may result from the level of change in disability required in this
analysis. These particular conditions may have more of an effect on changes in
functional limitations (as opposed to the tasks used here) and may present changes
in absolute ability that can be mitigated by adaptations in environment and
behavior. Alternately, the effects of these particular conditions could be
confounded with other characteristics in the model. While weight is not associated
with changes in functioning in the total sample, extreme low weight is associated
with an increased chance of decline for men. It would be useful to know whether
or not this low weight represents a long-term weight or a recent change. Changes
in body-mass ratio cannot be calculated from the data, however.
O f great interest is the effect of life-threatening diseases on sequence in
change in functioning. Among those included, heart disease, cerebrovascular
disease, and chronic obstructive pulmonary disease that reach a level of severity
requiring hospitalization during the course of the survey are each associated with a
significant risk of decline. However, when total Medicare cost is included, the
effects of heart disease and cerebrovascular disease disappear, which might suggest
that intensity of care can mitigate the disabling consequences of this condition.
168
However, increased utilization is associated with increased chances of decline as
well. Neither hospital episodes associated with malignant neoplasms nor a history
of cancer is associated with changes in functioning. This may be a result both of
the dynamics of the disease — that it is likely to either kill quickly or to become
controlled in such a way that the survivor does not suffer residual impairment —
and the chance that there is a relative lack of association between this pathology
and physical functioning. More curious is the lack of impairment resulting from a
history of a stroke or a cerebrovascular event. Because such events are likely to
repeat, this effect may be captured in the indicator for hospital episodes associated
with cerebrovascular disease.
A subset of the predictors of declines are associated with nonlinear
trajectories of change over the six years. In addition to these, this analysis adds
the new information (echoing other research on healthy behaviors) that exercise is
associated with maintenance of functioning.
Usage of hospital services may prevent death and may limit the extent of
functioning decline. However, the results from this study indicate that intensive
use of inpatient hospital services is associated with some decline in functioning and
is inversely related to maintenance of functioning over time. Obviously, persons
who are in poor health and thus more likely to decline will be more likely to use
health services. This analysis does not tell us how much worse-off those people
might have been without such hospital care.
169
These findings suggest that hospital stay medical costs in the older
population are being accumulated on tertiary care — responding to acute episodes
in such a way that a person is saved from death or perhaps more severe disability,
but nevertheless experiencing some decrement in functioning. However, in this
analysis types of care (curative, palliative, rehabilitative) were not differentiated.
Finally, there is a significant reservoir o f robustness in the older population. This
is the modal trend in trajectories. Increased emphasis in practice and research
should be on identifying the characteristics and processes that lead to maintenance
or resilience in functioning and working to bolster older persons’ good health.
170
Figure VI-1: Possible Transitions Between Levels of Functioning
No Difficulty
Unable in IADL , > Unable in ADL
Table_JJI^l£_Sam£le_Characi teristics_at_Baseline< _Interview> _£1984^i
Mean age
Mean years of education
Female
African American
In poverty
Not unable in ADL or IADL
Unable in IADL, not ADL
Unable in ADL
Full Those Who
Sample Lived and
Responded
at Every
Interview
(n=5,151)_______(n=2,430)
78 77
10 10
64% 67%
11% 11%
19% 17%
85% 93%
9% 5%
7% 3%
172
Table VI-2: Distribution of Functioning State at Four Interviews
(n=2,426)
Total Sample
Functioning State 1984 1986 1988 1990
N = Not unable in ADL/IADL 92.9 87.0 80.8 73.0
I = IADL disability only (no
ADL disability)
4.5 7.3 9.6 12.1
A = ADL disability 2.6 5.7 9.7 14.9
Men
Functioning State 1984 1986 1988 1990
N = Not unable in ADL/IADL 97.4 89.3 84.8 77.9
I = IADL disability only (no
ADL disability)
3.3 6.6 7.3 9.5
A = ADL disability 2.4 4.1 7.9 12.6
Women
Functioning State 1984 1986 1988 1990
N = Not unable in ADL/IADL 92.2 85.8 78.7 70.3
I = IADL disability only (no
ADL disability)
5.1 7.3 9.5 11.1
A = ADL disability 2.8 6.9 11.9 18.6
X2
.11 .02 .001 .000
173
Table VI-3: Frequency and Percent Distribution of Patterns of Change in
Functioning Health Over Four Years fn=2426)
Patterns of No Change Patterns of Fluctuation
Patterni Frequency Percent Pattern Frequency Percent
AAAA 32 1.9 AAIA 2 0.9
IIII 17 1.0 AANA 1 0.4
NNNN 1637 97.1 AIAA 1 0.4
Total 1686 100.0 AIAN 1 0.4
Sample Percent 69.4 AINA 1 0.4
ANAA 5 2.1
ANA I 1 0.4
ANAN 1 0.4
ANIA 2 0.9
ANII 1 0.4
Patterns of Improvement ANNA 2 0.9
Pattern Frequency Percent ANNI 2 0.9
IAAI 2 0.9
AAAI 3 15.0 IAIA 4 1.7
AAAN 1 5.0 IAII 4 1.7
AAII 1 5.0 I ANA 1 0.4
AANN 2 10.0 IANI 1 0.4
ANNN 5 25.0 I ANN 1 0.4
IIIN 1 5.0 IINA 4 1.7
IINN 3 15.0 IINI 5 2.1
INNN 4 20.0 INAA 2 0.9
Total 20 100.0 I NAN 1 0.4
Sample Percent 0.9 INIA 1 0.4
INII 9 3.8
ININ 4 1.7
INNI 5 2.1
NAAI 1 0.4
NAAN 1 0.4
Patterns of Decline NAIA 3 1.3
Pattern Frequency Percent NAII 4 1.7
NAIN 2 0.9
IAAA 18 3.8 NANA 6 2.6
IIAA 12 2.5 NANI 1 0.4
IIIA 9 1.9 NANN 11 4.7
NAAA 37 7.7 NIAI 8 3.4
NIAA 21 4.4 NIAN 1 0.4
NIIA 11 2.3 NUN 11 4.7
NIII 21 4.4 NINA 5 2.1
NNAA 51 10.6 NINI 18 7.7
NNIA 37 7.7 NINN 26 11.1
NNII 49 10.2 NNAI 17 7.3
NNNA 91 19.0 NNAN 17 7.3
NNNI 123 25.6 NNIN 38 16.2
Total 480 100.0 Total 234 100.0
Sample Percent 20.0 Sample Percent 9.8
r = Not unable in ADL or IADL
; = unable to do at least one IADL
A = Unable to do at least one ADL
174
Table VI-4: Percentage Distribution by Sex* of Types of Change in
Functioning State
Men Women Total
Improving 0.9 0.7 0.8
Declining 16.4 21.9 20.1
Fluctuating 8.2 10.3 9.6
Constant 74.6 67.0 69.5
Total 100.0 100.0 100.0
(798) (1623) (2421)
175
Table VI-5: Percentage Improving, Declining, Fluctuating and Staying at the Same Level of Functioning for
Each of Three Intervals and Over the Three Intervals___________________________________________________
1984=>1986 1986=>1988 1988=>1990 Pattern
Improving
Declining
Constant
Fluctuating
Total
2.0%
(48)
9.1%
( 2 2 1 )
89.0%
(2152)
100. 0%
(2421)
4.3%
(105)
11.7%
(283)
84.0%
(2033)
1 0 0. 0%
(2421)
4.4%
(107)
13.7%
(332)
81.9%
(1982)
1 0 0. 0%
(2421)
0 .8%
(19)
2 0. 1%
(487)
9.6%
(232)
69.5%
(1683)
100%
(2421)
o\
TabljB_VI^6jD£aj^terns==ofD_Chanc[e_and_Fun£tionincj_Status__i _i ____^_j __i __
Patterns of Functioning Change by Initial Functioning State (1984)
Pattern ADL
Disability
IADL
Disability
No Disability
Constant 50.00 15.74 72.82
Improvement 18.75 7.41
------
Decline
------
36.11 19.62
Fluctuation 31.25 41.67 7.56
Total 100.00 100.00 100.00
N 64 108 2248
Pattern of Functioning Change by Final Functioning State (1990)
Pattern ADL
Disability
IADL
Disability
No
Disability
Constant 8.91 5.80 92.59
Improvement
------
1.37 0.90
Decline 79.94 65.87
------
Fluctuation 11.14 26.96 6.50
Total 100.00 100.00 100.00
N 359 293 1768
177
Table VI-7: E ffects' of Sociodem ographic, Economic, Health, and Health Care Utilization Characteristics on
Uje_Conditional_Oddsj>yhtavin|y|j|«mernj>fjCharKy|jniiF i ^
PHArl?n«/Prnr» Prf«rlinn /P/»nn P flttrt/P m n
Intercept 0.000 0.000 0.000 0.000
Age 1.115 1.121 1.075 1.077
Female 2.141 2.162 .
African American(non-AA) . .
Years of education . .
Belou poverty threshold • .
missing income - .
Heart disease hosp. stay 1.800 .
Malignant neoplasm hosp. stay . .
Cerebrovascular disease hosp. stay 2.484 1.885 .
Chronic obstructive pulmonary disease 7.153 5.856 4.151 3.509
Heart disease a t baseline 0.660 0.609 .
Cancer a t baseline . . .
Stroke or cv event a t baseline . . .
Diabetes 1.928 2.084 .
A rth ritis . . .
Osteoporosis • . •
D ifficu lty hearing . . .
D ifficu lty with vision 6.334 6.631 4.350 4.489
Low 10% body-mass ra tio 1.818 1.805 .
High 10% body-mass ra tio . . .
Ualks a mile 3x per week . . 0.563 0.555
IA D L d is a b ility 9 . . 7.211 6.737
A D L d is a b ility 9 N.A. N.A. . .
Number of impairments 1.392 1.385 1.434 1.417
Changed respondents 8.607 8.134 4.019 3.863
Total Medicare Part A costs N.A. 1.034 N.A. 1.024
N o Medicare record match N.A. 1.625 N.A. .
Likelihood Ratio9 Chi-sq 2318.97 2266.00 2468.40 2427.76
df 3956 4062 4052 4158
B
1 . 0 1 . 0 1 . 0 1 . 0
Chi-sq
df
E
N
model 2 v model 1
52.97
4
.0001
2086
model 2 v model 1
53.92
4
.0001
2135
9 The omitted category is having no A D L or IA D L d is a b ility .
b This is a te s t of the specified model compared to a saturated model; nonsignificance indicates
the specified model is appropriate. ' All odds ra tio s presented are sig n ifican t a t p<.05
178
Table VI-8: Effects' of Sociodemographic, Economic, Health, and Health Care Utilization Characteristics on
the Conditional Odds of Havina a Pattern of Chanae in Functionina Characterized as Decline, bv Sex
Women Men
Pdflcline/Pcon Pdeclino/Pco Pdecline/Pco
Intercept 0.000 0.000 0.000 0.000
Age 1.135 1.141 1.082 1.084
African American(non-AA) . . .
Years of education . 0.932 0.933
Below poverty threshold
■ . .
missing income . . .
Heart disease hosp. stay 2.098 . .
Malignant neoplasm hosp. stay ■ . .
Cerebrovascular disease hosp. stay 2.241 3.210 3.067
Chronic obstructive pulmonary disease 4.820 3.787 17.258 13.323
Heart disease a t baseline 0.437 0.394 . .
Cancer a t baseline . . .
Stroke or cv event a t baseline . 2.870 3.251
Diabetes 2.171 2.213 . .
A rth ritis . . .
Osteoporosis .
..." ..."
D ifficu lty hearing . . .
D ifficu lty with vision 5.301 5.869 14.531 15.007
Low 10% body-mass ra tio . 2.234 2.423
High 10% body-mass ra tio . . .
Walks a mile 3x per week . . .
IA D L d isab ility " . 0.171 0.151
Number of impairments 1.461 1.444 . .
Changed respondents 10.204 9.096 6.602 6.722
Total Medicare Part A costs N.A. 1.040 N.A. 1.026
N o Medicare record match N.A. . N.A. 2.625
Likelihood Ratio" Chi-sq 1534.21 1494.67 725.07 707.07
df 2634 2720 1277 1293
e
1.0 1.0 1.0 1.0
model 2 v model 1 model 2 v model 1
Chi-sq
df
E
N
52.82
4
.0001
1407
18.00
4
.01
679
The omitted category is having no A D L or IA D L d is a b ility .
b This is a te s t of the specified model compared to a saturated model; nonsignificance indicates
the specified model is appropriate.
All odds ra tio s presented are sig n ifican t at £<-05
Because no men had osteoporosis in th is selected sample, i t could not be included.
179
Table VI-9: E ffects' of Sociodem ographic, Economic, Health, and Health Care Utilization C haracteristics on
thejC onditional^O dds^ofJfavjngjgjyiltem jj^C hm igejn^Functjoning^C ^
Women
. Men
■Rllugt/Pcon PfliiRf/Prnrt PlliiRt/Prnn P flnnt/P rnn
Intercept
Age
African American(non-AA)
Years of education
Below poverty threshold
missing income
Heart disease hosp. stay
Malignant neoplasm hosp. stay
Cerebrovascular disease hosp. stay
Chronic obstructive pulmonary disease
Heart disease a t baseline
Cancer a t baseline
Stroke or cv event a t baseline
D i abetes
A rth ritis
Osteoporosis
D ifficu lty hearing
D ifficu lty with vision
L ow 10% body-mass ra tio
High 10% body-mass ra tio
Walks a m ile 3x per week
IA D L d is a b ility ”
0.000
1.094
1.795
0.000
1.009
4.548
15.112
4.969
0.260
12.466
Chi-sq
df
E
N
44.33
4
.0001
1442
0.000
29 157
6.732
0.000
21.693
16.103
A D L d is a b ility ” • ■ 23.336 25.641
Number of impairments 1.452 1.456 1.366 1.366
Changed respondents 3.842 3.539 5.761 5.785
Total Medicare Part A costs N.A. 1.034 N.A. .
N o Medicare record match N.A. . N.A. .
Likelihoodb Ratio Chi-sq 1610.07 1578.37 792.76 774.23
df 2702 2788 1305 1321
E
1.0 1.0 1.0 1.0
model 2 v model 1 model 2 v model 1
18.52
4
.001
694
" The omitted category is having no A D L or IA D L d is a b ility .
b This is a te s t of the specified model compared to a saturated model; nonsignificance indicates
the specified model is appropriate. ‘ All odds ra tio s presented are sig n ifican t a t p<.05 ” Because
no men had osteoporosis in th is selected sample, i t could not be included.
1 8 0
Table VI-10: Predicted Mean Probability of Declining or
Specific Values of Significant Independent Variables
Characteristics as Observed__________________________
Variable Value
Age
Age 70
Age 90
Sex
Female
Male
Cerebrovascular disease(hosp. stay)
Yes
No
Chronic obstructive pulmonary disease
(hosp. stay)
Yes
No
Decline
0.1242
0.3797
0.2215
0.1533
0.2631
0.1925
0.3944
0.1945
Heart disease (at baseline)
Yes
No
Diabetes
Yes
No
Vision
Has difficulty
No difficulty
Lowest decile body-mass ratio
Yes
No
Walks a mile 3x per week
Yes
No
Has IADL disability at baseline
Yes
No
Impairments
Difficulty with none
Difficulty with 1
Difficulty with 3
Difficulty with 6
Changed respondent
Yes
No
Total Medicare cost
None
§1,000
$10,000
$50,000
$100,000
Missing Medicare match
Yes
No
0.1645
0.2041
0.2739
0.1922
0.4039
0.1868
0.2711
0.1909
0.1656
0.1922
0.2506
0.3423
0.3595
0.1165
0.1676
0.1704
0.1965
0.3396
0.5471
0.2447
0.1918
Fluctuating for
with All Other
Fluctuate
0.0712
0.1130
0.1259
0.0834
0.1430
0.0807
0.0629
0.0968
0.2778
0.0739
0.0606
0.0741
0.1076
0.1748
0.1161
0.0717
0.0769
0.0777
0.0851
0.1202
0.1581
181
VII. PROCESSES OF CHANGE IN FUNCTIONAL HEALTH
A. Introduction
In developing the conceptual model of change in functioning health, the
notion that the effects of health events may accumulate with age in such a way that
future changes in health are dependent on previous changes in health was
introduced. This may be because wear on the system produces cumulative effects,
or because previous experience may indicate something about the process of
change in functioning ability that is not otherwise captured by hypothesized
influences. An example is illustrative: Consider the case of two individuals over
70 years of age. One has for at least four years had no functioning impairment
(perhaps never has) and currently has no functioning impairment. The other has
had a series of adverse health events and had been unable in IADL, but has
recently experienced some recuperation and is currently reporting no difficulty with
any of these activities. Does having no functioning difficulty mean the same thing
to these two individuals? Or, is the risk that the person who has never experienced
disability in ADL and IADL will become disabled in the next interval of
observation the same as the risk for the person who has had considerable
fluctuation in ability but currently happens to be without impairment?
Analysis presented in the literature implies a prevailing methodological
assumption that if the other predictors of functioning change are adequately
specified an individual’s status at the beginning of the interval in question will
appropriately capture all that needs to be known about previous functioning status.
182
This is to say, that while the history of predictors may matter, the history of their
effects does not. An alternate approach would be to postulate that changes in
functional status affect the probability of future changes, all other things remaining
the same.
The implication of the prevailing assumption is that previous functioning
change (or lack of change) does not affect future change. In this chapter, analyses
are presented that aim to test the assumption that functioning status at the
beginning of an interval sufficiently measures whatever needs to be known about
the previous change in functioning when predicting future change. Two
approaches are taken. First, a formal test of whether or not functioning change
conforms to a simple Markov process states more explicitly the assumptions of the
theory implicit in the literature. It also gives a sense of how closely this
hypothetical process comes to reflecting the observed changes in functioning.
Second, within the context of a multivariate model, the effects of including
information on previous functioning ability are examined. This gives a more direct
indication of the success of the models currently being employed. The strategy
proposed here provides considerable detail regarding the kinds of changes in
functioning an individual is likely to have, given their immediate history of
functioning change. If these tests suggest that previous functioning does matter,
conclusions regarding how previous loss of functioning will affect the chances of
future losses will be possible.
183
B. Test of a Simple Markov Model
The Markovian assumption is that if the initial distribution of a sample
among statuses and the rate at which individuals move to other states in an interval
are known, then the probability of an individual in any initial state ending up in
any other state after some number of transitions is also known. This is the basic
assumption implicit in the choice to reflect prior experience simply as status at the
beginning of the interval.
The sample members who remain alive through 1990 and respond to ADL
and IADL questions are distributed among three statuses in 1984: being able to do
all ADL and IADL tasks, being unable to do at least one IADL, and being unable
to do at least one ADL. Table V II-1 presents the proportions making transitions to
and from each of these states. In each interval, for each initial status, there is a
clear tendency to remain at the same level of functioning, as evidenced by the
largest percentages occurring along the diagonal. Over the three intervals,
however, it becomes less likely that a person will remain without disability, and
more likely that he or she will remain in the most disabled state. Among those
with no disability, it is less likely that ADL ability will be lost than IADL ability.
On the other hand, those with ADL disability are more likely to regain both ADL
and IADL ability than they are to regain only IADL ability until the third interval.
Those who are unable to do IADL tasks are about equally likely to regain that
ability as they are to lose ADL ability. These proportions are fairly constant
across the three intervals.
To test whether or not the transitions between statuses follow this same
pattern over time, the proportions changing status in the first interval are applied to
the resulting distribution at the second interval. This results in a projected
distribution at the third. Those initial proportions making specific transitions are
applied to the constructed third wave distribution to produce an expected final
distribution at the fourth interview. Table VII-2 compares the expected
distributions at the third and fourth interviews under this assumption with the
observed distribution. The difference between the predicted and actual
distributions are considerable, with the predicted distribution being healthier than is
actually observed. This discrepancy may suggest that change in functioning ability
does not follow a simple Markovian process. This may be an overly simplistic
test, however. The sample for this experiment was selected to include only
persons for whom ADL and IADL information was available at each of the four
interviews, and this artificially restricts both the population and the destination
states. In addition to staying alive and having some level of disability at the end of
an interval, a person could also die. Both not having included death as a
competing risk, and not compensating for nonresponse, should make the "actual"
distribution look healthier than the "expected." As it turns out, this is not the case.
The subsequent sections examine the contribution of information prior to initial
status (status at the beginning of the interval) on predictions of future change
within the context of a multivariate model.
185
C. Multivariate Analysis of Specific Transitions
A conventional analysis of transitions in functioning status is adopted to test
the adequacy of the assumption that status at the beginning of an interval of
observation captures what we need to know about the process of change in
functioning. Focusing on individual transitions (changes between two adjacent
interviews) allows information from prior interviews to be used in predictive
variables. In the analysis of patterns of change presented in the previous chapter,
the approach was to include all of the information on functioning change in the
dependent variable. While that approach permitted an examination of actual
individual trajectories, the typology necessary to summarize those trajectories
obscured, to some extent, the quality of the variation. In other words, while
direction of change was known, the level of change was not represented. The
approach taken in this chapter overcomes this limitation. Also, since this approach
focuses on transitions over an interval rather than patterns over six years, the
results from these two approaches can be contrasted to illuminate how various
characteristics affect functioning change over the long-term and the short-term.
1. Sample
In order to maximize the number of transitions observed, data are pooled
over intervals1. This analysis requires response on the interview prior to the
1 As discussed earlier (Ch. IV, footnote 2, page 74), this data structure can lead
to unobserved heterogeneity. In these analyses, such heterogeneity in propensity to
make transitions between levels of functioning ability might be expected to arise from
person-specific differences in underlying frailty. Information on both prior functional
186
initial interview of the interval of interest and the two interviews defining that
interval. Therefore, the sample is limited to persons who have at least three
consecutive completed interviews. The observations include those who respond to
interviews in 1984, 1986, and 1988, and those who respond to interviews in 1986,
1988, and 1990. Those who respond to all four interviews contribute two person-
interval observations to the sample. Because there was no nonresponse in 1984, all
persons in this sample contribute a third interval observation (1988-1990). The
sample contributing observations for the second interval (1986-1988) includes all of
those individuals, and also some persons who responded to the three requisite
interviews but then became nonrespondant or deceased by the 1990 interview.
First interval transitions (1984-1986) are not analyzed because prior functional
status is not known. Because the data are pooled over intervals, an indicator of
whether or not the observation comes from the second interval (reference category
is the third interval) is included in all analyses to control for the effect of the
interval. This control variable is not significant in any o f these analyses.
2. Measures
The dependent variables in the transition analyses indicate whether or not a
transition was made from a given state at the beginning of an interval to another
specified state at the end of the interval. The dependent variables are dichotomous
status and on precursors to functional disability are included in these models,
however, and these variables are expected to capture any such underlying health
differences not otherwise reflected in the models.
187
indicators of whether status remains the same, or changes to a specific other status.
For each possible initial and end status, a separate model is estimated. The
possible statuses are the same as those used for the dependent variable in the
previous chapter in the analysis of patterns of change (Chapter VI). The rationale
for these states is developed in Chapter V. These states are: Able to perform each
of the ADL and IADL tasks, unable to perform IADL tasks (but able to perform
ADL tasks), and unable to perform ADL tasks. Thus, dependent variables indicate
whether or not a transition was made from no disability to IADL disability,
whether or not a transition was made from no disability to ADL disability, whether
or not a transition was made from IADL disability to no disability, and so on.
The predictive model is identical to that used in the analysis of patterns of
change in the previous chapter (Chapter VI), as developed in Chapter II
(Conceptual Framework). In addition to being the model closest to the theoretical
conceptualization that the data will allow, it also provides comparability between
the two analyses. The advantage of this is that differences and similarities in the
results can be interpreted as differences or similarities in the construction of the
dependent variable. A parallel analysis in the subsequent section enhances this
comparability further.
In addition to those characteristics, these models also include indicators of
functioning status at the interview prior to the interval in question. These
indicators test the assumption that history of functioning ability does not matter.
188
3. Analysis
Transitions in functioning status between specific levels of ability are
analyzed. For each possible transition, a multivariate logistic regression is
estimated on the subsample eligible to make the transition of interest. Two models
are estimated for each transition. The first is the basic multivariate model
proposed in the previous chapter. In addition to the other variables of interest, this
model controls for functioning status at the beginning of the interval, because each
analysis is conducted on a subsample of initial functioning status. This model
represents the prevailing assumption in analyses of this kind. The second model
includes all of the same indicators and adds to them dummy variables indicating
the individual’s functioning status at the interview prior to the interval of interest.
An alternative way to represent prior experience would be to include
indicators for the type of transition in the prior interval. For example, indicators
that improvement or decline had occurred could be included. However, status at
the interview prior to the interval of interest is used instead because this controls
for the starting level at that prior interval, and implies the type of transition made
to reach the status at the beginning of the interval in question. This is more
information than an indicator of transition type would provide.
Figure V II-1 provides a visual aid to this analysis. Three of the six
transitions analyzed are depicted in this figure. The three columns of statuses
indicate prior status, and the initial and final status for the interval of interest (see
bottom of diagram). Interview dates for the observation of status appears across
189
the top of the diagram. The first sequence diagramed depicts the transition from
not unable to unable in IADL. In this case, dummy variables will indicate the
effect of ADL or IADL inability (relative to ability) in the previous interview on
the odds of making a transition from being able to being unable in IADL (relative
to the odds of remaining able in IADL).
In the prevailing theoretical model, the previous status is implicitly
irrelevant, and indicators of previous status would be expected to be insignificant.
In the alternative model proposed here, previous experience does matter. In the
example presented in the introduction to this chapter (and depicted in the first
diagram in Figure VII-1), persons who are previously not unable would be
expected to be more likely to remain not unable in the interval of interest
compared with those who were previously unable in IADL or unable in ADL and
then became not unable. In this example, it is hypothesized that those persons are
more likely to become unable in IADL than to remain free of disability.
4. Predictors of Transitions to Worse Functioning by Initial Status
Table VII-3 presents the odds ratios for the effects of the hypothesized
covariates on each possible transition to worse health. Decline from able in all
tasks to IADL disabled is the most common transition, followed by decline from
able to ADL disabled, and then decline from IADL disabled to ADL disabled.
The actual number o f persons making each type of transition are not so different,
but the number eligible to make the transitions is considerably different, with 4,208
beginning with no disability and only 382 eligible to decline from IADL disability.
190
Factors Affecting Loss of IADL Ability Among Those With No Disability in ADL
and IADL
In the first model of the first analysis, older age, being female, being
African American, having a hospital episode associated with a primary diagnosis of
cerebrovascular disease or chronic obstructive pulmonary disease, having had
diabetes within the last year, having trouble with vision or hearing, having changed
respondent, and having more Medicare expenses or not having information to
match the respondent with Medicare claims files all increase the chance of a move
to IADL disability among those with no disability. Walking a mile at least three
times each week reduces this chance. Women are about 50% more likely than
men to have a decline in functioning relative to maintaining the level of No
Disability, and African Americans experience about that same risk compared to
non-African Americans. Some researchers have argued that the effect of sex and
race are associated with differences in access to health care (either lacking ability
to pay or discrimination in allocation of resources or use of procedures). In this
analysis, actual Medicare Part A usage has been controlled in dollars spent, and
the inability to match sample persons with their Medicare records is also
controlled. Even when these factors are taken into account, both women and
African Americans are more likely to make adverse transitions in functioning.
This may reflect differences in non-Medicare Part A health care access or
utilization. It may also reflect actual differences in disease processes not
adequately captured in other aspects of the model.
Among the various medical incidents and conditions, having had a hospital
stay with a primary diagnosis o f cerebrovascular disease, or having a history of
diabetes, each increase the relative risk of decline by about 60%. Hospital
episodes associated with chronic obstructive pulmonary disease and hearing
difficulty double that risk; difficulty seeing triples it. Those who have changed
respondent are almost four times as likely to have a decline. Each additional
thousand-dollar increase in Medicare in-patient hospital expenditure is associated
with an incremental increase in chance of decline. Those who were not able to
provide adequate information for matching to Medicare records are also those with
a worse prognosis. And, among these persons who are free of ADL and IADL
disability, each additional functioning limitation a person has at the beginning of
the interval increases the chance of decline by 30% compared to the chance of
remaining free of disability.
In the second model of the first analysis, dummy variables indicating
previous functioning status are included. The omitted category is No Disability.
W hile having had ADL disability in the previous interview does not affect the
outcome (IADL Disabled or No Disability), having had IADL disability at the
previous interview increases the chance of ending up with IADL impairment by
over four times. Thus, those who move from IADL Disability to No Disability are
more likely to return to IADL Disability by the next interview than they are to stay
at No Disability, controlling for all of the other hypothesized covariates.
192
Including the indicators of prior experience does not affect most of the
other predictors, but there are exceptions. The effects of a hospital episode
associated with chronic obstructive pulmonary disease and the beneficial effects of
regular exercise become insignificant when prior status is included. Alternately,
having had a hospital episode associated with heart disease becomes significant,
and is associated with a 30% increase in the relative risk of decline. There is a
modest shift in the effects of sensory impairment, but otherwise the effects are
remarkably robust.
Factors Affecting Loss of ADL Ability Among Those With No Disability in ADL
and IADL
In the second analysis, movement to ADL impairment during a two year
interval is examined for those who report no disability in ADL or IADL tasks.
The first model is similar to the corresponding model in the first analysis in terms
of salient predictors and the magnitude of effect, with a few exceptions. For
declines to this more intense level of dependency, the effects of sex and race are
not significant. Also, having a history of diabetes, or vision or hearing impairment
are also not significant predictors of this transition. The relative sizes of effects
for the other predictors are similar to those of in the models predicting decline to
IADL disability, although in most cases the effects are considerably stronger
predicting the more serious decline.
In the second model, analyzing transition from No Disability to ADL
Disability, having earlier disability prior to return of functioning considerably
193
increases the chances of subsequent decline. Previously having had ADL disability
increases the chance for loss of IADL functions among those who have no
disability two and a half times, and prior ADL disability has an effect twice that.
Factors Affecting Loss o f ADL Ability Among Those With IADL Disability
Surprisingly few persons begin an interval with IADL impairment and end
in either the same status or with ADL disability (n=382). Perhaps because of the
relatively small sample and large number of predictor variables, very few effects
are significant. Unique to this particular transition, extreme overweight is
associated with a reduced chance of decline in functioning2. Other salient
predictors of moves between these two states of disability include older age, being
female, and having difficulty seeing. Although number of limitations increases the
odds by about one-fifth, previous functioning is not a significant predictor of such
a change.
It is important to note the differences in the patterns of significance of
characteristics affecting the different transitions, as well as the relatively small
numbers o f people in the IADL disabled category experiencing decline in this
interval (and for whom information on the previous interval is available). These
2 W hile this relationship exists at a non-significant level in bivariate analysis, it
becomes significant only when the number of impairments, diabetes, and osteoporosis
are controlled. This unusual relationship should be understood as an artifact of this
highly selected sample until further evidence supports such an association more
generally.
194
low numbers may account for the lack of significant associations between many of
the predictor variables and this type of transitions in ability.
5. Predictors of Transitions to Better Functioning bv Initial Status
Although in general, relatively few people make transitions to improved
functioning, among those with IADL disability who either maintain that level of
functioning or improve, over a half of the persons make the transition to improved
functioning, the rest remain IADL impaired (Table VII-4). The chances of
improved functioning are reduced with increasing age, for those with a greater
number of limitations, among those who have had hospital episodes with a primary
diagnosis of malignant neoplasm, who have difficulty seeing, and who walk
regularly. Knowledge of previous functioning health does not improve the ability
to predict whether or not the change will occur.
Very small cell sizes resulting from the few transitions from ADL disability
to no disability and the relatively large number of covariates, precluded estimating
the chances of this kind of improvement. It was, however, possible to examine
recovery of ADL ability into IADL disability status. It is interesting that the
salient predictors for this model are quite different from those predicting recovery
to full functioning from IADL impairment. In the analysis of movement from
ADL disability to IADL disability, persons who have diabetes and who are
overweight are more likely.to regain some functioning. These results do not
indicate that it is beneficial to have diabetes or to be overweight — rather, they
suggest that persons with these conditions may gain and lose ability to do ADLs.
195
That is to say, that while the dynamics of their functioning health may involve loss
o f ADL ability, it is not the permanent or longer term loss that may occur with
other disease processes or with other constellations of conditions.
Previous research indicates that persons with higher education have, on
average, better health and a greater chance of maintaining functioning over time.
In this analysis, education has an effect opposite of what might be expected. As
with the results discussed above, this also indicates a selection related to the type
o f process that results in ADL disability.
When previous health status is included in the model, those who had ADL
disability in the prior interview are less likely than those who had no disability and
then lost ADL ability to regain functioning to the level of IADL impairment.
Whether, in that prior interview, an individual reports IADL impairment or no
disabilities is insignificant.
D. Multivariate Analysis of Types of Transitions
In this section a similar transition analysis is presented, but in contrast to
the previous section which focused on transitions between specific states, in this
analysis all transitions that can be characterized as "declines" are analyzed
together, and those that represent "improvement" are analyzed together. This
strategy allows an indicator of initial status (status at the beginning of the interval
over which the transition occurs) to be included, so that the effect of a given status
on propensity for either improvement, decline, or maintenance of functioning may
be evaluated.
196
The analysis of patterns of change in functioning ability (Chapter VI) was
conducted on a typology indicating whether improvements, declines, maintenance,
or fluctuation in ability were reported, and not on the specific levels of the
improvements and declines. This analysis uses a dependent variable that
characterizes changes generally as improvements or declines, and provides an
outcome that can be more readily compared to the analysis of patterns presented in
Chapter VI. An important difference is that change over a two-year period is now
being categorized as improvement or decline, whereas in the analysis of patterns,
this change was constrained to describe a six-year period. (Please refer to Chapter
VI, Table VI-5 for a comparison of the proportions changing under these two
definitions.)
The sample for the decline analysis is the combination of subsamples for the
analyses presented in Table VII-3, and the sample for the improvement analysis is
equivalent to the subsamples for the analyses presented in Table VII-4. Since this
analysis is not estimated for subsamples by initial status, but by those eligible to
make the type of transition of interest (rather than a specific transition), the
indicator for functioning status at the beginning of the interval is included. In the
analysis of declines the reference category for the indicator of initial functioning
status is No Disability, and in the analysis of improvement the reference category
is Unable in ADL. As in the previous analysis, a second model tests for the
effects of functioning ability prior to the interval in question. Previous experience
197
is again indicated by dummy variables for disability in ADL or IADL, and the
omitted category No Disability.
1. Predictors of Decline Over an Interval
Persons who are older, female, or have incomes below the poverty
threshold are more likely to experience declines over an interval, than others. The
number of years of education an older person has does not affect their chances of
decline over a two year period. Among health indicators, having a hospital
episode with a primary diagnosis of cerebrovascular disease, chronic obstructive
pulmonary disease, having had a stroke or cerebrovascular disease prior to the
initial interview in 1984, diabetes, or trouble hearing increase the chance of decline
in function. Neither current hospitalization associated with malignant neoplasm or
heart disease, nor history of cancer or heart disease are associated with declines in
this sample of survivors. Persons with diabetes and those who have difficulty
hearing are more likely than those without these conditions to report declines over
an interval, but arthritis, osteoporosis, and difficulty seeing are not associated with
transitions to worse health over two years. The chance of experiencing a loss of
ability is reduced by about one-third for those who walk regularly; weight does not
have a significant association with this type of change. And, those who change
respondent, have higher Medicare expenditures or are missing information for a
match with Medicare records are also more likely to experience a decline. These
characteristics are associated with an increased chance of decline over a two-year
period, controlling for the initial status. Initial functioning status, however, is not
significant, but those with a greater number of impairments are more likely to
experience a decline in functioning than to remain the same.
When previous level of ability is controlled, initial status becomes
significant. Similar to its effect on patterns of decline (Chapter VI), those with
IADL disability are less likely than those without disability to decline. And, both
indicators of prior experience have significant and considerable impact on the odds
of decline. Those whose functioning was previously impaired are over two times
as likely to experience a decline in functioning compared to maintaining the level
of functioning they had at the beginning of the interval, relative to those who
previously had no disability in ADL and IADL. With the exception of poverty
status, the effects of the rest of the indicators are stable. With previous health
experience included, the effect of poverty status becomes insignificant.
2. Predictors of Improvement Over an Interval
In the analysis of improvements over a two year interval, increased age
reduces the odds of improvement relative to staying at the same level of
functioning. Osteoporosis and trouble with vision also impede chances of
recovery. Those who change respondents are typically not those who experience
recovery, and those who did not have adequate information to match their
Medicare records are also less likely to be the ones recovering from impaired
functioning. Initial status is not significant in this model, and when previous status
is included, the effect of health conditions also disappear.
199
E. Summary and Comparison of Models
The goal of the analyses in this chapter was to determine whether or not it
is appropriate to use initial status for the interval of interest as the control for
previous functional health experience. To test this assumption, pairs of models
with and without additional information on previous functioning were estimated,
first for specific transitions, and then, to approximate the analysis of patterns in the
previous chapter, for types of transitions.
1. Models Predicting Decline
There were some similarities and some differences in the models predicting
decline (Table VII-3 and Table VII-5). Overall, decline between levels of
disability (IADL to ADL) presents a much different pattern than decline from
being able to being either unable in ADL or unable in IADL. These models share
only the effects o f age and number of impairments. While the decline from IADL
to ADL is associated with difficulty with vision and missing a Medicare record
match, the effects are opposite of those for the other specific transitions.
Transitions from No Disability to IADL Disability, and from No disability
to ADL Disability are predicted by similar factors: age, hospital episode for
cerebrovascular disease, hospital episode for chronic obstructive pulmonary
disease, not exercising regularly, having changed respondent, having higher
Medicare Part A expenses or not having a Medicare record match. There are
differences, however. Being African American is associated with higher risk of
transitions from no disability to IADL disability, as is having had a heart disease
200
related hospital episode, having diabetes, and having difficulty with one’s vision or
hearing.
In the analysis of transition from IADL to ADL disability, neither indicator
of previous functioning is significant. Previous IADL disability is an important
predictor of subsequent declines out of the no difficulty category — either to IADL
disability or ADL disability. Previously having had ADL disability is only
important for transitions from no disability to ADL disability.
The pattern of significance in the model predicting any type of decline
(Table VII-5) encompasses the significant predictors from both the model
predicting moves from No Disability to ADL Disability and from No Disability to
IADL Disability. The differences are that race is significant, and difficulty with
vision is not.
2. Models Predicting Improvement
Again, there is little similarity between the models predicting improvement
between categories of disability and improvement to no disability. Age is
significant in predicting returns to complete functioning from IADL disability, but
not for regaining only ADL ability (to the level of IADL disability). Similarly,
having had a hospital episode associated with a primary diagnosis of malignant
neoplasm, having difficulty with vision, or having a missing Medicare match, each
reduce the chance that a person will improve from IADL disability to having no
disability. Yet, they are each unimportant for predicting improvement from ADL
disability to IADL disability. That transition is made more likely by having had
201
diabetes and being overweight, and reduced by having changed respondents. It is
also more likely among those with lower levels of education. Previously having
had ADL disability reduces the chances of this type of improvement.
For the analysis predicting any type of transition to improved status over a
two-year interval, previous ADL or IADL disability limits the chances of regaining
functioning. Initial status is not significant. The effect of osteoporosis is unique to
this model predicting general improvement over a two year interval. Other
specific conditions such as malignant neoplasm, diabetes, and overweight are not
significant here.
F. Discussion
The analysis of specific transitions is informative in several ways.
Considering the specific transitions (compared to maintenance of level of
functioning/disability) allows a comparison of the differential effects of the
hypothesized covariates. The transition approach itself (considering change
between only two observation points rather than trajectories characterized over four
observations), permits the previous functioning experience to be used as predictive
information for future change. Findings from each of these angles are discussed
below.
1. Differences in Transitions by Level of Change
Analyzing transitions between each pair of functioning statuses allows both
direction and level of transition to be taken into account. Similarities and
differences suggest that movement between levels of disability are different from
202
movement in and out of no disability. For example, in the analysis of decline, the
variables in the proposed model are for the most part unrelated to the transition
between IADL disability and ADL disability. Although there are more similarities
between the models predicting loss of either ADL or IADL ability from having no
disability, there are interesting differences. Being African American, having been
hospitalized for heart disease, having a history of diabetes, having trouble hearing,
and having difficulty seeing (even with glasses) are each associated with loss of
IADL function, but not with loss of ADL function. And, although previous IADL
disability is an important predictor of future decline to either IADL or ADL
disability among those currently without disability, among those without disability,
previous ADL disability is only important in predicting return to that level of
impairment.
Similarly, old age, being hospitalized for malignant neoplasm, and having
difficulty seeing are important factors limiting recovery of full functioning among
those with IADL disability. Among those with ADL disability, none of those
characteristics affect the transition to IADL disability. And, for this transition, the
curious effects of higher education reducing the chance of improvement and
diabetes and being overweight increasing the chance of improvement suggest a
complicated selection effect associated with the type of disease processes that are
responsible for people ending up with ADL disability. For example, we know
from previous studies that higher education is frequently found to be associated
with better health by a variety of measures, and with a number of kinds of
203
those persons with higher education, while in general maintaining better health for
a longer period of time, when they do finally lose functioning ability, are likely to
do so in a permanent way. Also, it may be characteristic of the morbid processes
associated with diabetes and being overweight that there are changes in
functioning, including loss — but then recovery — of ADL ability. However, the
effects on recovery of functioning should be regarded cautiously because of rather
thin data for these transitions. And, previous ADL disability (which in this model
is indicator of duration of time with ADL disability) reduces the chances of
recovery of ADL ability. Neither previous ADL or IADL disability affects the
propensity to regain IADL ability.
2. Effect of Previous Experience on Subsequent Change
The second way this strategy is informative is that it permits the inclusion
o f information on previous functioning ability (and by inference, on previous
changes in functioning). In more cases than not, previous functioning experience
is an important predictor of future change.
Previous functioning does not impact the odds of moving from IADL
disability to ADL disability, or the odds of moving from IADL disability to no
disability. The lack of effect in these particular cases may be the relatively equal
"distances" of each of the previous status categories from IADL disability.
Previous functioning is an important predictor for other transitions.
Previous functioning is measured as status at the interview prior to the interval of
interest. The interpretation of previous status therefore varies with each specific
204
transition. In the first transition to worse health, moving from no disability to
IADL disability, having a previous IADL increase the odds of decline relative to
maintaining all abilities by four and a half times. This means that those people
who had at least one IADL, then regained functioning, are over four times as
likely to return to the level o f IADL impairment than are those who had no
disability at one interview, and also report no disability at the next. Those people
are likely to still have no disability at the end of the interval of interest.
Alternately, having had ADL impairment previously and then returning to a level
of no disability does not affect the chances of a subsequent decline. It may be that
persons who report themselves unable in one or more ADL at one interview, and
then able to do all ADL and IADL tasks two years later are persons who had an
acute health event, an accident, for example, from which they fully recover, and
that does not impact future functioning. In contrast, having IADL disability from
which one recovers may be indicative of a chronic disease process, that may be at
some times more or less manageable, but from which one never "fully recovers."
Thus, returns to IADL disability are likely.
Movement from no disability to ADL disability is influenced by any
previous disability. This means that among those persons who have no disability at
the beginning of an interval, those persons who improved to the level of no
disability, from either having had disability in ADL or IADL, are more likely to
experience loss of functioning again compared to those who consistently reported
having no disability for two interviews. It is interesting to note that, while
205
previous ADL disability did not affect chances of losing IADL functions, it does
affect the chances of losing ADL ability. Thus, it may be that the scenario of an
acute incident was incorrect, that it only holds for a subsample, or that it is
correct, but that the acute incident is repeatable or is associated with other types of
events from which one recovers fully within a two year period.
Both those who are unable to do ADL tasks and then become unable to do
IADL tasks, and those who are unable to do IADL tasks and then at a subsequent
interview are still unable in IADL tasks are equally as likely as those who have no
difficulty with any tasks and then become unable to do IADL functions to
subsequently report no disability. In other words, among those who are unable to
do one or more IADL tasks (but not any ADL tasks), previous functioning status
does not influence the chance of recovery of the IADL abilities. However, those
individuals who report being unable to do ADL tasks for two consecutive
interviews are two-thirds less likely to regain ADL abilities (recover to the level of
IADL disabled) than are those who had no difficulty and then became unable to in
ADL. This has two implications. First, the longer a person has a severe
disability, the less likely he or she is to recover functioning. Second, there are
apparently some morbid processes that result in rather quick loss of ability (from
no disability to ADL disability in two years) from which at least partial recovery is
possible. Because of a limited sample, it is not possible to see this in the transition
to complete recovery (ADL disability to no disability).
206
3. Conclusion
The objective of this chapter was to determine whether or not previous
functioning experience has an impact on future transitions that is not captured in
the individual’s status at the beginning of the interval in which the transition is to
be observed. Clearly the indicators of previous functioning add information to the
analysis of most of the transitions. Does this mean that initial functioning status at
the interval of interest inadequately captures the health history of the individual
when the goal is to predict change in functioning? Within the context of the
current model specification, it does. There are two possible interpretations of this,
however. It may be that this model is underspecified. If the theoretical model was
correct and perfectly measured, perhaps the initial status would summarize
everything that needed to be known about where in the process of disablement a
person is. Alternately, it may be that an individual’s history of maintenance or
change in functioning affects future ability to maintain or recover functioning or
the propensity to lose functioning in a way that is not indicated by history of
exogenous influences on functioning. The process of change in functioning may be
a cumulative process in which history matters. The analysis presented here does
not discriminate between these two interpretations. It does indicate that either
interpretation is possible, and therefore that further investigation is required. Even
if the results are finally found to indicate that models of changes in functioning
such as these are not adequately specified, at least the inclusion of previous
207
functioning experience can serve as a gross indicator for some of the omitted
characteristics until these characteristics are known and appropriately included.
208
209
Figure VII-1: Representation of Selected Processes of Change
1984
1986
Not Unable
Unable IA DL
Unable ADL
Prior Status
1986
1988
^ Not Unable
Unable IA DL Unable IA DL
Unable ADL
Not Unable
Unable ADL ► Unable ADL
Initial Status
I _____
1988
1990
Not Unable
^ Unable IADL
-► Unable ADL
Final Status
Interval of Interest
Table VII-1: Percent in Each Initial Status Making a Transition to Each Final Status for Every Interval
n=2,420
Interval 1 = 1984-1986
Interval 2 = 1986-1988
Interval 3 = 1988-1990
Initial Status Final Status
I No IADL ADL Total
Disability Disability Disability % N
1 91.6 5.4 2.9 100% 2,248
No Disability 2 88.8 6.7 4.5 100% 2,105
3 86.4 7.9 5.7 100% 1,955
1 24.1 47.2 28.7 100% 108
IADL Disability 2 35.2 39.8 25.0 100% 176
3 24.2 45.9 29.9 100% 231
1 29.7 4.7 65.6 100% 64
ADL Disability 2 17.3 14.4 68.4 100% 139
3 9.8 13.7 76.5 100% 234
Table VII-2: Predicted and Actual Prevalence of Disability
______ Expected______________ Actual
1988
No Disability 83.1% 80.8%
IADL Disability 8.5% 9.6%
ADL Disability 8.4% 9.7%
Total 100% 100%
1990
No Disability 80.6% 73.1%
IADL Disability 8.9% 12.1%
ADL Disability 10.4% 14.8%
Total 100% 100%
211
TabTeVII^3^0^dsRaJia|= ^£E= ^j^D3^i£gD£_E£S^i£52£5= ^£TransitTonsto^orseH eaT thb^lnit1aT Status
No = > IADL No = > ADL IADL = > ADL
Model 1 Model 2 Model 1 Model 2 Model 1 Model 2
Intercept 0.000 0.000 0.000 0.000 0.001 0.001
Previous IA D L
. . .
4.552
. . .
2.486
. . .
.
Previous A D L
. . .
.
. . .
4.421
. . .
.
Number of impairments 1.336 1.319 1.345 1.321 1.241 1.237
Age 1.097 1.097 1.113 1.109 1.060 1.062
Female 1.506 1.489
A frican American (non- 1.573 1.630
Education .
Under poverty threshold •
Hissing poverty ■
Hosp. heart disease 1.362
Hosp. malignant • 0.498
Hosp. cerebrovascular 1.609 1.643 2.147 2.087
Hosp. chr. obstructive 2.377 . 3.939 4.084
Baseline heart disease . .
Baseline cancer . .
Baseline stroke/cv . .
Diabetes past year 1.672 1.628
A rth ritis past year
• ■
Osteoporosis past year . ■
Hearing trouble 1.976 2.064
Vision trouble 3.007 2.425 0.380 0.366
Low 10% body-mass • •
High 10% body-mass . ■ 0.482 0.479
Walks mile a 3x / week 0.705 . 0.580 0.600
Changed respondent 3.893 3.732 5.245 5.206
Total Medicare cost 1.011 1.011 1.020 1.020
Missing Medicare match 1.604 1.587 1.599 1.534 0.510 0.498
Interval 2 . . . .
-2 Log L 2032.676 2000.381 1536.384 1517.564 442.925 442.561
covars ml v m 2 covars ml v m 2 covars ml v m 2
Chi-sq 523.873 32.295 489.760 18.82 76.532 .324
df 26 2 26 2 26 2
P
.0001 .0001 .0001 .0001 .0001 n .s.
response categories
*p < .05
1=377 0=3930 1=274 0=3930 1=160 0=222
212
Table VII-4: Odds Ratios for Significant* Predictors of Transitions to Improved Health by In itia l
Status_____________________________________________________________________________________
T A D I => N n A D I => N n A D I => T A D I
Model 1 Model 2 Model Model 2 Model 1 Model 2
Intercept 999.000 999.000 ■
Previous IA D L
. . .
.
. . .
■
Previous A D L
. . .
.
. . .
0.366
Number of impairments 0.817 0.822 0.630 0.650
Age 0.903 0.907
Female
African American (non-AA)
Education 0.916 0.909
Under poverty threshold
Hissing poverty indicator
Hosp. heart disease
Hosp. malignant neoplasm 0.232 0.228
Hosp. cerebrovascular
Hosp. chr. obstructive
Baseline heart disease
Baseline cancer
Baseline stroke/cv
Diabetes past year 2.408 2.572
A rth ritis past year
Osteoporosis past year
Hearing trouble
Vision trouble 0.323 0.353
Low 10% body-mass
High 10% body-mass 2.600 2.647
Ualks mile a 3x / week 0.410 0.408
Changed respondent 0.467 0.432
Total Medicare cost
Missing Medicare match 0.492
Interval 2
-2 Log L 402.894 401.094 378.484 368.773
covars ml v m 2 covars ml v m 2 covars ml v m 2
Chi-sq 70.597 1.8 60.987 9.711
df 26 2 26 2
P
.0001 .30 .0062 .01
response categories 1=135 0=222 1=52 0=367 1=86 0=365
p < .05
213
TableVIM a^Odd^Ratiosfor^SignjJicjmtlPredictorsofGeneralizedTransltions
Decline ___ Improvement
M odel 1 M odel 2 M odel 1 ■ MndeL2.-
Intercept
In itia l sta tu s IA D L *
Previous sta tu s IA D Lb
Previous sta tu s A D L "
Number of impairments
Age
Female
A frican American (non-AA)
Education
Under poverty threshold
Hissing poverty indicator
Hosp. heart disease
Hosp.. malignant neoplasm
Hosp. cerebrovascular d is.
Hosp. chr. obstructive
Baseline heart disease
Baseline cancer
Baseline stroke/cv
Diabetes past year
A rth ritis past year
Osteoporosis past year
Hearing trouble
Vision trouble
Low 10% body-mass
High 10% body-mass
Ualks mile > 3x / week
Changed respondent
Total Medicare cost
Hissing Medicare match
-2 Log L
Chi-sq
df
E
response categories
0.000
1.350
1.101
1.288
1.293
1.700
2.178
1.744
1.676
1.685
0.691
3.777
1.016
1.386
3392.86
covars
1026.99
26
.0001
1=811
0.000
0.691
2.135
2.037
1.339
1.099
1.291
1.718
2.201
1.614
1.674
1.621
0.699
3.803
1.016
1.338
3370.23
m l v m2
22.63
2
.0001
0=4,152
79.106
0.760
0.957
0.461
0.623
0.656
0.643
990.15
covars
84.72
26
.0001
1=273
118.999
0.677
0.248
0.780
0.956
0.619
959.92
m l v m2
30.23
2
.0001
0=5897
* In decline analysis reference category is Not Unable in A D L or IADL; In improvement analysis
reference category is Unable in A D L
b Reference category is Not Unable in A D L or IA D L
' p < .05
214
VIII. SUMMARY AND DISCUSSION
Loss and recovery of physical abilities occurs throughout older age, and
involves both gradual and relatively sudden change. In this study, four
observations of a panel of persons 70 years of age and older provide more explicit
detail on how socioeconomic, disease, health condition, and health behavior
indicators affect ability than has previously been reported from this type of data.
A combination of longitudinal approaches reveals: (a) those who had previously
experienced some functioning loss (even when ability is fully recovered) are more
vulnerable to subsequent loss of functioning than those who have not had previous
disability; (b) the number of physical limitations a person has can be used as an
indicator of risk of further decline in self-care and independent living abilities; (c)
diseases and other health conditions vary in their effects on functioning — some are
likely to result in declines and debilitation over the six-year period, while others
tend to be associated with shorter periods of disability because of recovery or
adaptation; (d) while the effects of socioeconomic and demographic characteristics
are largely subsumed under the more immediate health characteristics included in
the analyses, there are clear differences in propensity to gain or lose ability among
population subgroups, and these warrant further investigation.
These findings regarding precursors to disability, and the effects of various
characteristics on the trajectory of disability, add a new dimension to the study of
change in functioning ability, and suggest a shift in the conceptualization of change
in functioning from a monotonic transition to a highly variable and on-going
215
process. These elaborations improve our ability to predict risk o f decline in
functioning, and suggest points of intervention to waylay the onset o f disability.
Specific findings are discussed below in a summary organized around the
questions that have guided this study. This chapter concludes with
recommendations for future research.
A. Summary and Discussion of Findings
1. Results of Incorporating Information
from Multiple Observations
Two approaches provide different but complementary information on
change in functioning over multiple observations. In the patterns analysis, change
in functioning was defined by status at four times. That approach highlights the
diversity of actual individual trajectories. The typology necessary to summarize
those trajectories somewhat obscures the extent of the variation, however; while
direction of change was known, the level of change was not represented. Focusing
on individual transitions (changes between two adjacent interviews) allows
information from prior interviews to predict future change. And, with the
transition approach, differences in predictors of change between levels of ability
and between no disability and types of disability, can be observed.
(1) What patterns of change does a representative sample of older
persons experience over four, observation points?
216
(2) How do proportions of older persons improving, declining, or
maintaining functioning differ when the change is defined by
transitions rather than by patterns?
When functioning statuses at four points in time are linked together,
patterns are formed that represent a more detailed portrayal of changes and
stability in functioning than has been developed by past research. Sixty-six out of
a total of 81 possible patterns are observed in this sample. These patterns are
characterized as improving, declining, fluctuating, or remaining constant. Persons
in this sample are most likely to maintain their level of functioning, usually having
no disability (70%), than to experience a decline from which they do not recover
(20%), or a pattern of fluctuation (10%). Scarcely one percent of this sample
experiences an improvement that is not accompanied by declines over this six- year
period.
This approach to studying changes in functioning differs from other
analyses of change, which typically consider transitions between two points in
time. The proportions improving, declining or remaining constant when change is
defined according to a transition between two time periods are compared to change
defined by a pattern (Table VI-5). When change is constrained over a six-year
interval, the chances of improvement are more than halved, and chances of
declines are more than doubled. The relative magnitude of these proportions
remains the same, however; people are most likely to maintain their initial level of
functioning.
217
The variability in experience reflected in these patterns suggests that
research that represents changes in functioning as transitions in status between two
times does not fully capture the dynamism of ability at older ages, and raises new
questions about what causes people to lose or regain ability.
(3) Does knowing something about the history of an individual’s
functioning change improve our knowledge or ability to predict their
future functioning change?
(a) Are predictors of change in functioning different
if change is defined as a trajectory over multiple
observations instead of as change between two times?
The categorization of the individual patterns of change as improvement,
decline, fluctuation, and no change is used as a dependent variable in a
multivariate analysis (from Table VI-7). The results of the decline analysis can be
compared with the analysis of transitions (from Table VII-5) to see if the predictors
for change represented as six-year patterns differ from change represented as two-
year transitions (Table VIII-1).
Although the pattern of significance in the two models is very similar, there
are some interesting differences. There are two probable sources of the difference
in effects between the analysis of transitions and the analysis of patterns. They
may result from analytic differences, or they, could be providing more detail about
the effects of the predictors on functioning change.
218
Those characteristics that have different effects on two year transitions than
the four year patterns are discussed in detail below. Many of the characteristics
have similar effects over two years as over four years. Those who are older,
women, had hospital episodes associated with cerebrovascular disease, chronic
obstructive pulmonary disease, or diabetes, who changed respondents, spent more
M edicare dollars, or were not matchable to a Medicare record, were more likely to
experience declines both over an interval and over a six-year period. For most of
these, the odds ratios are considerably larger predicting patterns than predicting
transitions. The different size of effects reflects the "dilution" of effects in the
pooled sample because of the increased number of observations in the denominator
of the odds ratio; alternatively, it may be a more accurate representation of effects.
The perspective depends on whether one wants to describe change over an interval
or change over six years.
(3) Does knowing something about the history of an individual’s
«
functioning change improve our knowledge or ability to predict their
future functioning change?
(b) Does prior change in functioning predict
subsequent change in functioning?
In the analysis o f transitions, information on an individual’s history of
functioning ability that was included in the dependent variable in the analysis of
patterns now becomes a predictor of subsequent change. Comparing models with
and without this additional information provides a test of the prevailing analytic
219
assumption that change in functioning ability is dependent only on one’s status at
the beginning of the interval of observation. In the generalized transitions (Table
VII-5), previous disability increases the chances of subsequent decline and reduces
the chances of improvement compared to previously having no disability. And,
among those beginning with no disability, loss of ADL or IADL ability is more
likely for those with previous disability (Table VII-3). Prior functioning does not
add additional information on improvements; rather, its effect is confounded with
the prior existence of impairments. In general, then, those who have previously
suffered a loss of functioning, even if they have recovered, are more vulnerable to
future loss.
2. Summary of Findings on Predictors of Functioning Change
The characteristics used to predict change in functioning are the same in all
analyses, regardless of how functioning change is defined. The characteristics
included are guided by the conceptual issues discussed in Chapter II. In addition
to being the model closest to the theoretical conceptualization that the data will
allow, using a similar model also provides comparability between the two analyses.
The advantage of this is that differences and similarities in the results can be
interpreted as differences or similarities in the construction of the dependent
variable or other differences in the analytic approach itself. Comparing results
from the several analyses provides a robust description of the predictors of
functioning change.
220
(4) Are the same characteristics associated with transitions to and
from different "levels" of disability (i.e. ADL and IADL)?
(5) When a complete model including social, economic,
demographic, disease, condition and impairment information is
employed, what is the picture we get of characteristics that affect
functioning health?
(6) Does the conceptual distinction between impairment and
disability hold analytically?
(7) What is the relationship between diseases and disability?
A comparison of the models of decline in both the patterns analysis and the
transitions analysis (Table VIII-2) facilitates some generalizations about the
characteristics that do and do not predict loss of functioning, as well as some
departures from these generalizations.
Two characteristics increased the chance that abilities will be lost in every
model considered: A greater number of limitations and older age. Women were
more likely to have changes characterized by declines in every analysis except for
transitions downward to ADL impairment. For these transitions, sex was not an
important predictor.
A "limitation" is the inability to perform component operations that indicate
less than ideal physical ability, but not necessarily inability to carry out tasks
required for independent living or personal care. In Verbrugge and Jette’s (1994)
model of disablement, limitations result from disease and physiological limitations,
221
but are conceptually prior to disability (for further discussion, please refer to
Chapter II, section B). Limitations in this study include difficulty waking a quarter
of a mile; walking up 10 steps without rest; standing or being on feet for about two
hours; sitting for about two hours; lifting or carrying 10 pounds; and stooping,
crouching or kneeling. Indicators of limitation are often omitted from analyses of
functioning change, or are included as a level of disability in the dependent
variable. In this analysis, additional information was gained by including an
indicator of the number of limitations in every model considered except for the
analysis of patterns of decline over six years among men.
Including a broad array of both health and socioeconomic characteristics
provides some insight into relationships between these characteristics and health
that are more ambiguous in studies that have their focus on specific sets of
predictors. In this study, race, education, and poverty status are not significant
predictors of change in functioning. In other studies, these characteristics have
been found to be highly correlated with health — persons who are African
American, have lower levels of education, or have incomes below the poverty
threshold are more likely to have worse health in research using a variety of
measures. The lack of association between these characteristics and change in
functioning in this study could be the result of several factors. First, if the
characteristics that define the socioeconomic differences work indirectly through
the health indicators, the socioeconomic and demographic indicators might not
themselves be significant in this analysis. This does not mean that there are not
222
differences in propensity to experience changes in functioning by these
characteristics -- only that the more proximate causes of the change, the health
conditions, have already captured that difference. Results not presented here
indicate that when socioeconomic status (education and poverty status) and the
more proximate health indicators are not controlled, African Americans are more
likely to have declines in functioning. When education and poverty status are
controlled, race loses its independent effect. Finally, when the disease and
condition indicators are included, the effects of socioeconomic status largely
disappear. These findings echo the work of House et al. (1994), which uses
indicators of psychosocial stress rather than health conditions to predict change in
functioning.
The results from the present study indicate that while there are significant
differences in patterns of functioning change by race, education, and poverty
status, that these differences are first manifest in the presence of disease and other
health conditions. This finding has two implications. First, it suggests that
differences in reporting of functional status by race are not measurement error, but
rather reflect actual differences in health by race (African Americans compared to
all others); these health differences result in differences in functioning status.
Second, it directs our attention to the nature of these differences and their causes.
House et al. (1994) find evidence that race, education and income differences in
health behaviors (smoking, alcoholic drinking, and weight) lead to differences in
chronic disease and functional status reports. Those authors interpret these effects
as "psychosocial" consequences of the stress of low socioeconomic status. In
addition to these behavioral differences, differences in life-time health, access to
health care utilization or health care utilization behavior may also contribute to
differences in health at older ages. Although there is evidence that the pronounced
association between socioeconomic status and health is weaker among older
persons because the prevalence of poor health among those of higher
socioeconomic status increases to the levels among those with lower socioeconomic
status (House, et al., 1994), in this study of older persons, these effects are still
present in the absence of disease and condition indicators. How these differences
arise are clearly important questions for future research.
In addition to these substantive reasons for null findings associated with this
group of predictors, it is also possible that effects of race, education, and poverty
have been attenuated by the selection of the sample for only the persons who
survived and responded through 1990. This selection may have eliminated the
least healthy or it may have truncated the distribution of the characteristics (Chyba
& Fitti, 1991; Foley et al., 1990; Streib, 1982). Those in the worst health are
more likely to be excluded from this sample. And, although neither the simple
comparison of characteristics between the total baseline sample and the sample
selected for this study, nor the analysis of how nonresponse affects the prevalence
of disability, suggest that serious truncation of the distribution of these
characteristics has occurred, it is possible that the selection would alter the
association between these characteristics and health.
There are two important exceptions to the general pattern of nonsignificance
o f effects of race, education, and poverty status. Persons who are African
American and have no disability are significantly more likely than non-African
Americans to become IADL disabled over an interval, all other things equal.
Those who are below the poverty threshold are more likely to report a decline over
a two-year interval than those with more income. And, men who have more years
o f education are more likely to maintain their level o f functioning than to
experience decline over a six-year period. This is in contrast to the finding of Mor
et al. (1989), that functioning for women, but not men, is influenced by
educational attainment. This particular difference is probably the result of
constraining change to a six-year trajectory; although how and why men but not
women should benefit from increased years of education deserves further
exploration.
Among the diseases and conditions considered, chronic obstructive
pulmonary disease, cerebrovascular disease, and diabetes are clear targets for
prevention and intervention. Their effects on chance of decline among survivors
are strong and consistent. These are diseases that are likely to compromise active
life expectancy: they impede functioning, but are not associated with such a high
rate of mortality that persons suffering from these diseases are selected out of the
sample.
Other conditions provide more food for thought but less conclusive
implications. Heart disease and malignant neoplasms, for example, might be
225
associated with loss of function, but they are also associated with high rates of
mortality and as such, those suffering impaired functioning as a result of these
diseases are selected out of this sample. For those who do not die, the natural
history of these diseases is such that they do not continue to impact an individual’s
risk of functioning decline over time. Thus, among those who remain in the
sample, either no association is found with the disease, or, in the case of those
with a history of heart disease, a change in behavior apparently results in a long
term reduction in risk of functional decline. Having a history of stroke at the
beginning of the survey predicts decline across an interval, but does not affect the
longer pattern of change. Although stroke is usually associated with some loss of
functioning (accounting for the increased chance of decline over an interval), over
time, some of this functioning may be recovered. This would result in a pattern of
fluctuation, improvement, or no change, depending on where in the process the
interview first occurred.
Chronic musculoskeletal diseases (arthritis and osteoporosis) are among the
first conditions that come to mind when we think about conditions that cause
functioning impairment in old age. While these conditions limit functioning, they
do not account for change between levels of disability. Six years may not be long
enough to detect the long-term progression of these conditions. The null effects of
arthritis and osteoporosis suggest that the effects of these known disablers (a) occur
coincident with diagnosis and/or (b) occur over such a protracted period that even
among a large group of sufferers incidence of disability (or decline to another level
226
of disability) over a six-year period is relatively low, and/or (c) that changes in
functioning among those suffering with these musculoskeletal conditions occurs
within the coarse levels of disability/no disability employed in this study.
Among sensory impairments, it is interesting that difficulty hearing
demonstrates an effect over the short-term, but not the long-term, whereas vision
impairment is associated with long-term impairment more consistently than with
transitions between two points in time. The differences in these effects suggest
that people are able to adapt more readily to loss of hearing than loss of vision, or
that interventions are available to mitigate hearing loss more effectively than vision
loss.
Finally, weight and exercise are inconsistently, but not usually, associated
with risk of decline. Overweight at baseline is associated with a reduced chance of
progressing from IADL to ADL disability over an interval. This effect may
suggest the opportunity to reduce risk by reducing weight. Those who report
walking regularly are less likely to experience decline over an interval, but they
are not less likely to decline over the long-run. These findings for both exercise
and weight probably arise because these characteristics were only recorded at the
initial interview in 1984, and both are the sorts of characteristics that may change
over a six-year time period.
It is interesting that other health conditions almost never predict loss of
functioning. Having a hospital episode associated with heart disease, controlling
for severity/intensity of service utilization and history of heart disease along with
227
the other characteristics, is associated with an increased risk of decline only for
transitions from no disability to IADL disability. History of heart disease is
associated with decreased risk of decline in the patterns analysis only for the full
sample and for the subsample of women. Hospital episodes associated with a
primary diagnosis of malignant neoplasm or a history o f cancer at baseline are not
significant predictors of declines. The lack of effect among these characteristics is
probably a result of the selected sample and the natural history of these conditions.
Increased risk of decline for persons with cancer and heart disease may taper off
rather quickly as the sickest die and others recover. Since the experiences of those
who die before 1990 are not represented in this analysis, declines among those who
become deceased during this period are not included in the analysis.
Although there are differences in predictors of loss of ADL and IADL
disability among those with no disability, there are very few predictors of
movement between ADL and IADL disability. Women and African Americans
with no disability are more likely to lose IADL ability than men and non-African
Americans, but they are not more likely to lose ADL ability. Likewise, among
persons with no disability, those having diabetes or difficulty with vision or
hearing are more likely to lose IADL disability than are those without these
conditions. Having these conditions does not effect the risk of decline to ADL
disability for those with no disability. Walking a mile at least three times each
week reduces risk of ADL disability among those with no disability, but it does not
affect the risk of losing IADL ability once previous status is controlled. Although
228
a number of the hypothesized predictors are significantly associated with change
from no disability, this set of characteristics is not very informative on the matter
o f loss o f ADL ability once IADL ability is lost.
In these analyses, both characteristics from several historical time points
(e.g. baseline characteristics and subsequent health events, and functioning status at
the beginning o f an interval as well as at the interview prior to that) and attributes
that represent various stages in the disablement process (i.e. disease, impairment,
and disability) are used to predict either patterns or transitions in functioning.
Certainly some of these characteristics that are either historically or conceptually
prior to others could be used to predict other right-hand side variables. Such
endogeneity could produce confounded associations and biased estimates.
Exploration of the robustness of the effects under varying specifications suggests
that this has not occurred in these analyses. However, rigorous tests of
endogeneity have not been employed.
B. Significance of Findings and Implications for Future Research
The goal of this study was to test the common analytic assumption that an
individual’s history of functioning status does not affect future change. And,
having established that observations at multiple time points do increase our
understanding of change in functioning, a second goal was to present a more
detailed and accurate description of the process and determinants of change in
functioning health by incorporating more information on individual characteristics
and on each individual’s functioning status over multiple interviews.
229
The findings presented here indicate that previous functioning status or
change does indeed alter the chances of subsequent change in functioning ability.
This is an important finding because (a) it helps us pinpoint those most likely to
lose ability and to determine the long-run chances for those who have regained
functioning from a given level of disability, and also (b) because it suggests a shift
is needed in how change in functioning is conceptualized. Including information
on multiple observations of change requires a shift in how we think about change ~
from monotonic, implicitly unidirectional transitions, to sequences of changes that
are not independent of each other, and that may follow many patterns varying in
both the timing and degree of change. Introducing information on more than two
points in time makes the analytic representation of individual experience more
realistic. The method presented in the transition analysis also introduces
information that adds a considerable amount of additional detail on the process of
change.
A second area of contribution from this study comes from the added detail
about predictors of change in functioning that is provided by comparisons across
the varied models. In addition to the substantive findings discussed above, these
comparisons highlight several refinements in how to think about predictors of
functioning change, including differentiating types of change in disability associated
with different diseases and conditions, and the consequences of various choices in
how change is represented and what is captured in those representations.
230
These refinements are most evident in the way diseases and conditions are
variously significant or insignificant in this study. The lack of significance of
certain conditions such as malignant neoplasms and heart disease suggest that
researchers should develop an understanding about the natural history of various
diseases and conditions and incorporate this into the study of change in
functioning. In particular, knowing the duration of increased risk of mortality and
loss of functioning introduced by individual disease processes would be an
important contribution to our understanding of the association between morbidity,
mortality, and disability, and would inform research on determinants of active life
expectancy. This research takes an important step in this direction, indicating
through multiple comparisons how different diseases vary with respect to the
period of risk of functioning decline associated with them.
The effects of musculoskeletal disease on functioning are complex.
Verbrugge (1992) has found that persons with arthritis are more likely than those
without arthritis to regain ADL and IADL abilities. The null findings here suggest
either that the change associated with the presence of these conditions had already
occurred, and/or that subsequent change occurs relatively rarely over a period of
only six years. Studies that can observe the incidence of those diseases, as well as
their long-term properties, could shed light on which of these alternative
interpretations is most accurate. It is likely that considerable fluctuation in ability
or level of difficulty is occurring, however; probably these changes are taking
place within the gross categories of disability used in this study. Future studies
may be able to detect less extreme changes with a more detailed measure of
disability.
Medicare Part A-covered expenditures are consistently associated with
increased risk of decline. Presumably, those who are worse-off will be those with
hospital stays, but we might expect for these people to at least have patterns of
fluctuation and not outright decline; that is, hospital treatment might be expected to
restore some functioning. Higher expenditures may be associated with more
serious conditions or with more intensive service utilization. This may be an
important distinction, and a more refined measure could detect this.
M ore importantly, this effect suggests.that hospital use is providing largely
primary intervention - it is retaining people in the population at more impaired
levels of functioning. An effective approach to turning this relationship around is
the use of ADL and IADL batteries as measures of effectiveness of medical
interventions. This use of ADL and IADL items is relatively new, and has been
implemented largely according to a medical model that does not include controls
for the broader array of salient characteristics included in this analysis. Large
scale outcomes research using ADL and IADL measures in the context of social
and epidemiological models would considerably improve the utility of this nascent
stream of research.
The finding that persons who had not provided adequate information to
allow a Medicare record match (characteristic of the majority of those missing a
match), had a significant increase in probability of loss of functioning is a new and
232
disturbing finding. An unmatched record occurring when a person has provided
(what they believe to be) his or her Medicare health insurance claim (HIC) number
can result either because that number was incorrectly reported (or recorded) or
because the other pieces of information, such as name, social security number, or
Railroad Retirement Board number used to verify the match were incorrectly
reported or recorded. Because the missing match predicts declines in functioning,
it is unlikely that recording errors are a significant contribution; such errors would
be random and would not show this association. Errors specifically providing the
HIC number could indicate difficulty understanding the Medicare system and thus
difficulty accessing the system. If this were the case, we might conclude that
persons who are not effectively accessing their Medicare Part A benefits have a
greater chance of losing their ability to live independently or to provide basic
personal care. Such a finding would clearly have important policy implications.
Alternately, the unmatched cases may be occurring because of general errors in
reporting the several pieces of information used to make the records match. Loss
of cognitive functioning would be an obvious culprit for this sort of error. The
LSOA unfortunately lacks a good measure of cognitive functioning with which to
test this hypothesis. It would be useful to look for the association between
unmatched records and poor health in other data sets that do collect measures of
cognitive functioning. Although not included in this study, cognitive functioning
and mental health are likely contributors to functioning ability. The omission of
233
measures of cognitive functioning and mental health is not expected to have
affected other results in this study.
Predictors of fluctuation are similar, though not identical, to predictors of
decline. This suggests that some of those who present patterns of fluctuation might
otherwise be decliners — differing perhaps in compensatory behaviors,
rehabilitation or other treatments, or perhaps in other unobserved characteristics
associated with the cause of the changes in functioning. Periods of fluctuation
might also be embedded in overall patterns of decline. Or, they might be the
context within which acute health events, accompanied by loss and recovery of
function, occur. And, for some, this observed period of fluctuation might have
occurred within an overall quite stable longer-term trajectory. Our understanding
of the dynamics of ability and disability will also be enhanced as new data
spanning even longer periods of time provide more context for interpreting periods
of improvement, decline, and fluctuation. Also, comparing results from surveys
that observe functioning ability at intervals of different lengths will add information
on what happens between interviews that are spaced farther apart. This will help
refine interpretation of results based on fragmentary data, and will also suggest the
most efficient observation plans for future surveys.
In the analysis of patterns, very few people had patterns of change that
could be characterized as improvement. The resulting small sample precluded
analysis of improvements using that technique. Yet, even using the transition
approach, which included enough observations for several analyses of
234
improvement, few characteristics were found to be associated with improvement.
Mathiowetz and Lair (1994) have argued that the inability to predict improvement
indicates that observed improvement in functioning is actually measurement error
produced by unreliable responses. In the study presented here, the levels of
disability are defined in such as way that unreliability is minimized. There is a
dramatic difference in ability between being able to bathe oneself and not being
able to bathe oneself; one is not likely to make an erroneous report about this sort
of ability. Thus, it is very unlikely that the improvement measured here is sullied
by any considerable error. It is of utmost importance, now that we have recovered
from our surprise that older persons can indeed regain functioning, that we turn
our efforts to collecting adequate samples so that these relatively rare but desirable
transitions may be observed more frequently, and to discover what makes those
lucky few bounce-back.
235
Table VIXI-1: Comparison of Odds Ratios Predicting Decline Over an
Interval with Patterns of Decline___________________________________
Transition Pattern
Pdecline/Pconst Pdecline/Pconst
Age
Female
African American
Years of education
Below poverty threshold
missing income
Heart disease hosp. stay
Hospital malignant neoplasm
Hospital cerebrovascular
disease
Chronic obstructive pulmonary
disease hosp stay
Heart disease at baseline
Cancer at baseline
Stroke of cv event at baseline
Diabetes
Arthritis
Osteoporosis
Difficulty hearing
Difficulty with vision
Low 10% body-mass ratio
High 10% body-mass ratio
Walks a mile 3x per week
Number of impairments
Changed respondent
Total Medicare Part A costs in
thousands of dollars
1.099
1.291
1.718
2.201
1.614
1.674
1.685
0.699
1.339
3.803
1.016
1.121
2.163
1.885
5.856
0.609
2.084
6.631
1.805
1.385
8.135
1.034
No Medicare record match 1.338 1.625
236
Table VIII-2: Sum m ary of Patterns of Significance for All Models of Decline
Age
Female
African
American
Education
Poverty
missing
i ncome
Heart disease
hospital stay
Malignant
neoplasm
hospital stay
Cerebrovascular
d is. hosp. stay
Chronic obst
pulmonary
disease
hospital stay
Baseline heart
disease
Baseline cancer
Baseline
stroke/cv
Diabetes
A rth ritis
Osteoporosis
Hearing
Vision
Low 10% body-
mass
High 10% body-
mass
Ualks mile
3x/week
Number of
impairments
Changed
respondent
Total Medicare
Cost
N o Medicare
match
P attern Pattern P attern Trans Trans Trans Trans
Total M en W om en None=>IADL Hone=>AD L 1A D L = > A D L Decl.
+
+
N A
+ +
+ + + +
+ +
+ + + + + +
+ + + + + +
+ + + + + +
+ + + + . +
237
REFERENCES
Adams, M. M. E ., Scherr, P. A ., Branch, L. G ., Herbert, L. E ., Cook, N. R .,
Lane, A. M ., Brock, D. B., Evans, D. A ., & Taylor, J. O. (1990). A
comparison of elderly participants in a community survey with
nonparticipants. Public Health Reports. 105(61. 617-622.
Allen, S. M ., Mor, V ., Raveis, V ., & Houts, P. (1993). Measurement of Need
for Assistance with Daily Activities: Quantifying the Influence of Gender
Roles. Journal of Gerontology. 48(41. S204-211.
Allison, P. D. (1984). Event History Analysis: Regression for Longitudinal Event
Data. Newbury Park: Sage.
Allison, P. D. (1994). Using panel data to estimate the effects of events.
Sociological Methods and Research. 23(2), 174-199.
Angel, R. J., Angel, J. L., & Himes, C. L., (1992). Minority group status, health
transitions, and community living arrangements among the elderly.
Research on Aging. 14(41. 496-522.
Antonucci, T. C. (1990). Attachment, social support, and coping with negative life
events. In E. M. Cummings, A. L. Greene, & K. H. Karraker, Life Span
Developmental Psychology: Vol. 11. Stress and Coping Across the Life
Span. Hillsdale, N .J.: Lawrence Erlbaum.
Berkman, L ., Singer, B., & Manton, K. (1989). Black/White differences in health
status and mortality among the elderly. Demography. 26(4), 661-678.
Boult, C., Kane, R. L., Louis, T. A., Boult, L., & McCaffery, D. (1994).
Chronic conditions that lead to functional limitation in the elderly. Journals
of Gerontology. 49(1), M28-M36.
Branch, L. (1985). Health practices and incident disability among the elderly.
American Journal of Public Health. 75(12), 1436-1439.
Branch, L., Katz, S., Kniepmann, K., & Papsidero, J. A. (1984). A prospective
study of functional status among community elders. American Journal of
Public Health. 74(3), 266-268.
Branch, L., & Ku, L. (1989). Transition probabilities to dependency,
institutionalization, and death among the elderly over a decade. Journal of
Aging and Health. 1(3), 370-408.
238
Burtless, G. (1987). Occupational effects on health and income among the elderly.
In G. Burtless (Ed.), Work. Health and Income Among the Elderly (pp.
103-150). Washington, D .C .: Brookings Institution.
Carter, R ., Coast, J. R., & Idell, S. (1992). Exercise training in patients with
chronic obstructive pulmonary disease. Medicine and Science in Sports and
Exercise. 24(3), 281-291.
Chyba, M. M ., & Fitti, J. E. (1991). Response experience in a longitudinal study
o f older persons. Proceedings of the American Statistical Association:
Section on Survey Research Methods. 135-140.
Corder, L. S., & Manton, K. G. (1991). National surveys and the health and
functioning of the elderly: The effects of design and content. Journal of the
American Statistical Association. 86(4141. 513-525.
Corder, L. S., Woodbury, M ., & Manton, K. G. (1992). Loss to follow-up
assessment: Application of grade of membership methods to aged
population longitudinal sample loss to follow-up in the National Long Term
Care Survey. Proceedings of American Statistical Association: Section on
Survey Research Methods. 357-361.
Crimmins, E. M ., Hayward, M. D ., & Saito, Y. (1994). Changing mortality and
morbidity rates and the health status and life expectancy of the older
population. Demography. 31(1), 159-175.
Crimmins, E. M ., & Saito, Y. (1993). Getting better and getting worse:
Transitions in functional status among older Americans. Journal of Aging
and Health. 5(1), 3-36.
Cunningham, W. R. (1991). Issues in factorial invariance. In L. M. Collins & J.
L. Horn (Eds.) Best Methods for the Analysis of Change: Recent Advances.
Unanswered Questions. Future Directions, (pp. 106-113). Washington, D.
C.: American Psychological Association.
Dowd, J. J., & Bengtson, V. L. (1978). Aging in minority populations: An
examination of the double jeopardy hypothesis. Journal of Gerontology.
33(3), 427-436.
Downes, J. (1952). The longitudinal study of families as a method of research.
Milbank Memorial Foundation Quarterly Bulletin. 30, 101-118.
239
Ferraro, K. F. (1987). Double jeopardy to health for black older adults? Journal of
Gerontology. 42(5), 528-533.
Ferraro, K. F. (1993). Are black older adults health-pessimistic? The Journal of
Health and Social Behavior. 34(3), 201-14.
Fitti, J. E ., & Kovar, M. G. (1987). A multi-mode longitudinal study of aging.
Proceedings of the American Statistical Association Section on Survey
Research Methods.
Foley, D. J., Branch, L. G ., Madans, J. H ., Brock, D. B., Guralnik, J. M ., &
Williams, T. F. (1990). Physical Function. In J. C. Coroni-Huntley, R. R.
Huntley, & J. J. Feldman (Eds.), Health Status and Well-being of the
Elderly: National Health and Nutrition Examination Survev-I Epidemiologic
Follow-up Study (pp. 221-236). New York: Oxford University Press.
Fried, L. P ., Ettinger, W. H ., Lind, B., Newman, A. B., & Gardin, J. (1994).
Physical disability in older adults: A physiological approach. Journal of
Clinical Epidemiology. 47. (7), 747-760.
Gerety, M. B., Mulrow, C. D ., Tuley, M. R., Hazuda, H. P ., Lichtenstein, M.
J., Bohannon, R., Kanten, D. N ., O ’Neil, M. B., Gorton, A. (1993).
Development and validation of a physical performance instrument for the
functionally impaired elderly: The physical disability index (PDI). Journal
of Gerontology. 48(2). M33-M38.
Gibson, R. C. (1991). Race and the Self-reported Health of Elderly People.
Journal of Gerontology. 46(5), S235-242.
Goldscheider. F. (1993). Does it make sense to talk about changes in family
structure? Presentation at the Andrus Gerontology Center Multidisciplinary
Colloquium Series, January 1993.
Guralnik, J. M ., & Kaplan, G. A. (1989). Predictors of healthy aging: Prospective
evidence from the Alameda County study. American Journal of Public
Health. 79(6), 703-708.
Guralnik, J. M ., & Simonsick, E. M. (1993). Physical disability in older
Americans. Journals of Gerontology. 48, 3-11.
240
Haber, L. D. (1990). Issues in the definition of disability and the use of disability
survey data. In Levine, D. B., Zitter, M. & Ingram L. (Eds.) Disability
statistics, and assessment. Committee on National Statistics, Commission on
Behavioral and Social Sciences and Education, National Research Council,
(pp. 35-51). Washington, D .C.: National Academy Press.
Hagestad, G. (1990). Social perspectives on the life course. In R.H. Binstock, &
George, L.K. (Eds.) Handbook of Aging and the Social Sciences. (3rd ed.)
(pp. 151-168). New York: Academic Press.
Harris, T ., Kovar, M. G ., Suzman, R., Kleinman, J. C ., & Feldman, J. J. (1989).
Longitudinal study of physical ability in the oldest-old. American Journal of
Public Health. 79(6), 68-72.
Herzog, A. R., & Rodgers, W. L. (1992). The use of survey methods in research
on older Americans. In R. B. Wallace, & R. F. Woolson (Eds.), The
Epidemiologic Study of the Elderly. New York: Oxford University Press.
Hill, M. S. (1992). The Panel Study of Income Dynamics: A user’s guide.
Newbury Park: Sage.
House, J. S., Lepkowski, J. M ., Kinney, A. M ., Mero, R. P ., Kessler, R. C.,
Herzog, A. R. (1994). The social stratification of aging and health. Journal
o f Health and Social Behavior. 35. 213-234.
Iannacchione, V. G ., Milne, J. G., & Folson, R. E. (1991). Response probability
weight adjustments using logistic regression. Proceedings of the American
Statistical Association: Section on Survey Research Methods. 637-642.
Jagger, C ., Clarke, M ., & Cook, A. J. (1989). Mental and physical health of
elderly people: Five year follow-up of a total population. Age and Ageing.
18, 77-82.
Jay, G. M ., Liang, J., Liu, X ., & Sugisawa, H. (1993). Patterns of nonresponse
in a national survey of elderly japanese. Journal of Gerontology. 48(3),
S143-S152.
Kalton, G. (1990). Comment on "Theory and issues in weighting survey data."
Proceedings of the American Statistical Association: Section on Survey
Research Methods. 147-149.
241
Kalton, G ., Kasprzyk, D ., & McMillen, D. B. (1989). Nonsampling errors in
panel surveys. In D. Kasprzyk, G. Duncan, G. Kalton, & M. P. Singh
(Eds.), Panel Surveys (pp. 249-270). New York: John Wiley & Sons.
Kalton, G ., & Kish, L. (1981). Two efficient random imputation procedures.
Proceedings of the American Statistical Association: Section on Survey
Research Methods. 146-151.
Kalton, G ., Lepkowski, J., & Lin, T.K. (1985). Compensating for wave
nonresponse in the 1979 ISDP Research Panel. Proceedings o f the
American Statistical Association: Section on Survey Research Methods.
372-377.
Kalton, G ., Lepkowski, J., Montanari, G. E., & Maligalig, D. (1990).
Characteristics of second wave nonrespondents in a panel survey.
Proceedings of the American Statistical Association: Section on Survey
Research Methods. 462-467.
Kalton, G ., & Miller, M. E. (1986). Effects of adjustments for wave nonresponse
on panel survey estimates. Proceedings of the American Statistical
Association: Section on Survey Research Methods. 194-199.
Kaplan, G. A ., Strawbridge, W. J., Camacho, T., Cohen, R. D. (1993). Factors
associated with change in physical functioning in the elderly: A six year
prospective study. Journal of Aging and Health. 5(1), 140-153.
Katz, S. (1987). The science of quality of life. Journal of Chronic Disease. 40(6),
459-463.
Katz, S. C ., Ford, A. B., Moskowitz, R. W ., Jackson, B. A., & Jaffee, M. W.
(1963). Studies of illness in the aged. The Index of ADL: A standardized
measure of biological and psychosocial function. Journal of the American
Medical Association. 185. 914-919.
Kasprzyk, D ., & McMillen, D. B. (1987). SIPP: Characteristics of the 1984 panel.
Proceedings of the American Statistical Association: Social Statistics
Section. 181-186.
Kempen, G. I. J. M ., & Suurmeijer, T. P. B. M. (1990). The development of a
hierarchical polychotomous ADL-IADL scale for noninstitutionalized elders.
The Gerontologist. 30(4), 497-502.
242
Kim, J., & Mueller, C. W. (1978). Factor Analysis: Statistical Methods and
Practical Issues. Beverly Hills: Sage.
Kish, L. (1990). Weighting: Why, when, and how? Proceedings of the American
Statistical Association: Section on Survey Research Methods. 121-130.
Kovar, M. G ., Fitti, J. E. & Chyba, M. M. (1992). The Longitudinal Study of
Aging: 1984-1990. Vital and Health Statistics. 1(28). Hyattsville: National
Center for Health Statistics.
Kovar, M. G ., & Feinleib, M. (1991). Older Americans present a double
challenge: Preventing disability and providing care. American Journal of
Public Health. 81(3), 287-288.
Kovar, M. G ., & Lawton, M. P. (1994). Functional Disability: Activities and
Instrumental Activities of Daily Living. In M. P. Lawton & J. A. Teresi
(Eds.), Annual Review of Gerontology and Geriatrics. 14. New York:
Springer, pp 57-75.
Kurtz, M. E ., Given, B., Kurtz, J. C., & Given, C. W. (1994). The interaction of
age, symptoms, and survival status on physical and mental health of patients
with cancer and their families. Cancer. 74(7), 2071-2078.
Laforge, R., Spector, W ., & Sternberg, J. (1992). The relationship of vision and
hearing impairment to one-year mortality and functional decline. Journal of
Aging and Health. 4(1), 126-148.
Lawton, M. P. (1982). Competence, environmental press, and the adaptation of
older people. In M. P. Lawton, P. G. Windley, & T. O. Byerts (Eds.),
Aging and the Environment: Theoretical Approaches. New York: Springer.
Lawton, M. P ., & Brody, E. M. (1969). Assessment of older people self-
maintaining and instrumental activities of daily living. The Gerontologist. 9,
179-185.
Lazaridis, E. N ., Rudberg, M. A., Furner, S. E., & Cassel, C. K. (1994). Do
activities of daily living have a hierarchical structure: an analysis using the
Longitudinal Study of Aging. Journals of Gerontology. 49(2), M47-51.
Lepkowski, J. M. (1989). Treatment of wave nonresponse in panel surveys. In D.
Kasprzyk, G. Duncan, G. Kalton, & M. P. Singh (Eds.), Panel Surveys
(pp. 348-374). New York: John Wiley & Sons.
243
Lepkowski, J. M ., Kalton, G ., & Kasprzyk, D. (1989). Weighting adjustments for
partial nonresponse in the 1984 SIPP panel. Proceedings of the American
Statistical Association: Section on Survey Research Methods. 296-301.
Lillard, L. A. (1989). Sample dynamics: Some behavioral issues. In D. Kasprzyk,
G. Duncan, G. Kalton, & M. P. Singh (Eds.), Panel Surveys (pp.
497-511). New York: John Wiley & Sons.
Lillard, L. A ., & Brien, M. J. (1993). Buying quality time: Wealth, health and
mortality. Presentation at the annual meetings o f the Gerontological Society
of America, New Orleans.
Little, R. J. A. (1986). Survey nonresponse adjustments for estimates of means.
International Statistical Review. 54(2), 139-157.
Little, R. J. A ., & Rubin, D. B. (1987). Statistical Analysis with Missing Data.
New York: Wiley.
Maddox, G. L., & Clark, D. O. (1992). Trajectories of functional impairment in
later life. Journal of Health and Social Behavior. 33, 114-125.
Magaziner, J. (1992). The use of proxy respondents in health studies of the aged.
In R. B. Wallace & R. F. Woolson (Eds.), The Epidemiological Study of
the Elderly. New York: Oxford University Press.
Manton, K. G. (1988). A longitudinal study of functional changes and mortality in
the United States. Journal of Gerontology. 43, S153-S161.
Manton, K ., Corder, L., & Stallard, E. (1993). Changes in the use of personal
assistance and special equipment from 1982 to 1989: Results from the 1982
and 1989 NLTCS. The Gerontologist. 33(2), 168-176.
Markides, K. S. (1989). Aging, gender, race/ethnicity, class and health: A
conceptual overview. In K. S. Markides (Ed.), Aging and Health.
Perspectives on Gender. Race. Ethnicity and Class. Newbury Park: Sage.
Mathiowetz, N. A ., & Lair, T. J. (1994). Getting better? Change or error in the
measurement of functional limitation. Journal of Economic and Social
Measurement. 20, 237r262.
Mihelic, A.H. & Crimmins, E.M . (1995) Loss to follow-up in a sample of
Americans 70 years of age and older in the LSOA: 1984-1990. In review.
244
Miller, D. C. (1991). Handbook of Research Design and Social Measurement. (5th
ed). Newbury Park: Sage.
M or, V ., Murphy, J., Masterson-Allen, S., Willey, C., Razmpour, A ., Jackson,
M. E ., Greer, D ., & Katz, S. (1989). Risk of functional decline among
well elders. Journal of Clinical Epidemiology. 49(9), 895-904.
Mor-Barak, M. E ., Miller, L. S. (1991). A longitudinal study of the causal
relationship between social networks and health of the poor frail elderly.
The Journal of Applied Gerontology. 10(3), 293-310.
Mor-Barak, M. E., Miller, L. S., Syme, L. S. (1991). Social networks, life
events, and health of the poor frail elderly: A longitudinal study of the
buffering versus the direct effect. Family Community Health. 14(2). 1-13.
Nagi, S. Z. (1991). Disability concepts revisited: implication for prevention. In
Pope, A. M ., & Tarlov, A. R. (Eds.) Disability in America: Toward a
national agenda for prevention. Division of Health Promotion and Disease
Prevention., Institute of Medicine, Washington, D .C .: National Academy
Press, pp 309-327.
Nakayama, H ., Jorgensen, H. S., Raaschou, H. O ., Olsen, T. S. (1994). The
influence of age on stroke outcome: The Copenhagen Stroke Study. Stroke.
25(4), 808-813.
Norris, F. H. (1985). Characteristics of older nonrespondents over five waves of a
panel study. Journal of Gerontology. 40(5), 627-636.
Palmore, E. B., Nowlin, J. B., & Wang, H. S. (1985). Predictors of function
among the old-old: A 10-year follow-up. Journal of Gerontology. 40(2),
244-250.
Pinsky, J. L ., Branch, L. G ., Jette, A. M ., Haynes, S. G., Feinleib, M ., Cornoni-
Huntley, J. C., & Bailey, K. R. (1985). Framingham Disability Study:
Relationship of disability to cardiovascular risk factors among persons free
of diagnosed cardiovascular disease. American Journal of Epidemiology.
122(4). 644-656.
Pinsky, J. L ., Leaverton, P. E., & Stokes, J. III. (1987). Predictors of good
function: The Framingham Study. Journal of Chronic Disability. 40(1),
159S-167S.
245
Posner, J. D ., McCully, K. K ., Landsberg, L. A ., Sands, L. P. Tycenski, P .,
Hofmann, M. T ., Wetterholt, & K. L ., Sands, L. P. (1995). Physical
determinants of independence in mature women. Archives of Physical
Medicine and Rehabilitation. 76(4), 373-380.
Reuben, D. B., Valle, L. A ., Hays, R. D ., & Siu, A. L. (1995). Measuring
physical function in community-dwelling older persons: A comparison of
self-administered, interviewer administered, and performance-based
measures. Journal of the American Geriatrics Society. 43(11. 17-23.
Rodgers, W. L ., & Herzog, A. R. (1992). Collecting data about the oldest old:
Problems and procedures. In R. M. Suzman, D. P. Willis, & K. G.
Manton (Eds.), The Oldest Old (pp. 135-156). New York: Oxford
University Press.
Rodgers, W. L., & Miller, B. (1994). A comparative analysis of ADL question
wording in surveys of older persons. Working paper from the AHEAD
Early Results Workshop. Ann Arbor.
Rogers, R. G ., Rogers, A ., & Belanger, A. (1992). Disability-free life among the
elderly in the United States. Journal of Aging and Health. 4(1), 19-42.
Roos, N. P., & Havens, B. (1991). Predictors of successful aging: A twelve year
study of Manitoba elderly. American Journal of Public Health. 81(1). 63-
68 .
Rosow, I., & Breslau, N. (1966). A Guttman Health Scale for the Aged. The
Journal of Gerontology. 21. 556-559.
Rowland, M. L., & Forthofer, R. N. (1993). Adjusting for nonresponse bias in a
health examination survey. Public Health Reports. 108(3’ ). 380-386.
Rubenstein, L. Z ., Schairer, C., Wieland, G. D ., & Kane, R. (1984). Systematic
biases in functional status assessment of elderly adults: Effects of different
data scores. Journal of Gerontology. 39(6), 686-691.
Rubin, G. S., Roche, K. B., Prasadarao, P., & Fried, L. P. (1994). Visual
impairment and disability in older adults. Optometry and Vision Science.
21(12), 750-760.
SAS Institute Inc. (1989). SAS/STAT User’s Guide. Version 6. Fourth Edition.
Volumes 1 and 2 . Cary, N .C .: SAS Institute Inc.
246
Schaie, K. W ., Labouvie, G. V., & Barrett, T. J. (1973). Selective attrition effects
in a fourteen-year study of adult intelligence. Journal of Gerontology.
28(3), 328-334.
Shryock, H. S., & Siegel, J. S. (1976). The Methods and Materials of
Demography. New York: Academic Press.
Siegler, I. C ., & Botwinick, J. (1979). A long-term longitudinal study of
intellectual ability of older adults: The matter of selective subject attrition.
Journal of Gerontology. 34(2), 242-245.
Simonsick, E. M ., Lafferty, M. E., Phillips, C. L ., Deleon, C. F. M ., Kasl, S.
V ., Seeman, T. E., Fillenbaum, G., Hebert, P., & Lemke, J. H. (1993).
Risk due to inactivity in physically capable older adults. American Journal
of Public Health. 83(10), 1443-1450.
Siu, A. L., Hays, R. D ., Ouslander, J. G ., Osterwell, D ., Valdex, R. B.,
Krynski, M ., & Gross, A. (1993). Measuring functioning and health in the
very old. Journal of Gerontology. 48(1), M10-M14.
Smith, D. S., & Clark, M. S. (1995). Competence and performance in Activities
of Daily Living of patients following rehabilitation from stroke. Disability
and Rehabilitation. 17(1), 15-23.
Soldo, B. (1985). In-home services for the dependent elderly: Determinants of
current use and implications for future demand. Research on Aging. 7, 280-
304.
Speare, A. J., Avery, R., & Lawton, L. (1991). Disability, residential mobility,
and changes in living arrangements. Journal of Gerontology. 4£(3),
S133-S142.
Spector, W. D ., Katz, S, Murphy, J. B., & Fulton, J. P. (1987). The hierarchical
relationship between Activities of Daily Living and Instrumental Activities
of Daily Living. Journal of Chronic Disability. 40(6): 481-489.
Strawbridge, W. J., Kaplan, G. A., Camacho, T. & Cohen, R. D. (1992). The
dynamics of disability and functional change in an elderly cohort: Results
from the Alameda County Study. Journal of the American Geriatrics
Society. 40(8), 799-806.
Streib, G. F. (1982). Participants and drop-outs in a longitudinal study. Journal of
Gerontology. 21, 200-209.
247
Teachman, J. (1982). Methodological issues in the analysis of family formation
and dissolution. Journal of M arriage and the Fam ily. 44, 1037-1052.
Vanschaardenburg, D ., Vandenbrande, K. J. S ., Ligthart, G. J., Breedveld, F. C .,
& Hazes, J. M . W. (1994). Musculoskeletal disorders and disability in
persons aged 85 and over: A community survey. Annals o f the Rheumatic
Diseases. 52(12), 807-811.
Verbrugge, L. M. (1992). Disability transitions for older persons with arthritis.
Journal o f Aging and Health. 4(2), 212-243.
Verbrugge, L. M ., & Jette, A. M . (1994). The disablement process. Social
Science and M edicine. 32(1), 1-14.
Verbrugge, L. M ., Reoma, J. M ., & Gruber-Baldini, A. L. (1994). Short-term
dynamics of disability and well-being. Journal o f Health and Social
Behavior. 35(2), 97-117.
Vuchinich, S., Teachman, J., & Crosby, L. (1991). Families and hazard rates that
change over time: some methodological issues in analyzing transitions.
Journal of M arriage and the Family. 53(4), 898-912.
Wolinsky, F. D ., & Johnson, R. J. (1991). The use of health services by older
adults. Journal o f Gerontology. 46(6), 345-357.
World Health Organization. (1976). The International Classification of
Impairments. Disabilities, and Handicaps. Geneva, Switzerland: W HO. In
Kovar & Lawton (1994).
World Health Organization. (1984). Uses of epidemiology in aging: Report of a
scientific group, 1983. Geneva, Switzerland: WHO Technical Report
Series. 706. WHO. In Katz (1987).
W right, R. A. (1983). Survey design and administration of the National Medical
Care Utilization and Expenditure Survey. Proceedings o f the 19th National
Meeting of the Public Health Conference on Records and Statistics
(DHHS-PHS84-1214), 177-180.
Yamaguchi, Kazuo (1991). Event History Analysis. Newbury Park: Sage.
248
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
An analysis of nonresponse in a sample of Americans 70 years of age and older in the longitudinal study on aging 1984-1990
PDF
The measurement of physical functioning in a longitudinal study of Americans 70 years of age and older
PDF
Mechanisms linking SES to hypertension: Findings from three longitudinal studies among the pre-retirement-age and older populations.
PDF
Ethnic differences in adult mortality in the United States: Evidence from a prospective study over 21 years
PDF
Infant mortality and its determinants in Los Angeles County, 1989-1991.
PDF
Evidence of age-based rationing of health care to the elderly: The effect of age and death on the use of health care services by elderly enrollees in an HMO
PDF
Secondary conditions and secondary disability among persons aging with long-term physical disability.
PDF
Procreative intentions in rural China: The relation with sex compositions of existing children, sex preferences, and social-economic status
PDF
Influence of physical activity, social support, and widowhood on disease incidence, mortality, and functional decline in older adults.
PDF
Legislative intent and aging policy in Los Angeles County: the relative effects of age versus need in public conservatorship
PDF
Socioeconomic status and cancer risk among diverse populations in Los Angeles County, 1972-1992
PDF
Elevated inflammation in late life: predictors and outcomes
PDF
Institutions and the search for security in old age: Fertility behavior, labor force participation, savings, insurance, and bequests in rural Thailand
PDF
Using self-reported memory ability to measure memory ability in older adults: A meaningful measure or an inapropriate shortcut?
PDF
Patient life in a nursing home: an experiential study
PDF
Studies of carotenoid and vitamin A complexes with protein in plasma and tissues
PDF
Futile medical treatment: an analysis of physician decision-making
PDF
A *longitudinal analysis of occupation in stroke recovery: A single case history
PDF
A prospective cohort study of morbidity and mortality among middle-aged men in Shanghai, China
PDF
Toward reconciliation of democratic and bureaucratic values in public professional organizations. Focusing on professional autonomy and organizational structures: The case of Korean public health...
Asset Metadata
Creator
Mihelic, Adrienne Heidt Pixley
(author)
Core Title
Processes of loss and recovery of physical abilities in a longitudinal study of Americans 70 years of age and older
Degree
Doctor of Philosophy
Degree Program
Gerontology
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Gerontology,health sciences, human development,OAI-PMH Harvest,sociology, demography
Language
English
Contributor
Digitized by ProQuest
(provenance)
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c17-80111
Unique identifier
UC11354926
Identifier
9625266.pdf (filename),usctheses-c17-80111 (legacy record id)
Legacy Identifier
9625266.pdf
Dmrecord
80111
Document Type
Dissertation
Rights
Mihelic, Adrienne Heidt Pixley
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA
Tags
health sciences, human development
sociology, demography