Blueprint for Healthy Aging, Theoretical Models and Concepts
bwilliams327
Multidomain.pdf
Multidomain Trajectories of Psychological Functioning in Old Age: A Longitudinal Perspective on (Uneven) Successful Aging
Jennifer Morack Pennsylvania State University
Nilam Ram Pennsylvania State University and Max Planck Institute for
Human Development, Berlin, Germany
Elizabeth B. Fauth Utah State University
Denis Gerstorf Pennsylvania State University and Humboldt University
Life-span developmentalists have long been interested in the nature of and the contributing factors to successful aging. Using variable-oriented approaches, research has revealed critical insights into the intricacies of human development and successful aging. In the present study, we opted instead for a more subgroup-oriented approach and examined multiple-indicator information of late-life change at the person level. We applied latent profile analysis to 8-year longitudinal data pooled together across 4 Swedish studies of the oldest old (N � 1,008; Mage � 81 years at Time 1; 61% women). Results revealed 4 psychosocial aging profiles with uneven patterns of successful (and less successful) aging characterized by distinct trajectories of change across indicators of depressive symptoms, social, and memory functions: a preserved system integrity group of participants who maintained functioning across very old age; an aging in isolation group with a persistent lack of social support, and 2 groups of people with average well-being and social functions but distinctive memory profiles. A compromised memory group was characterized by poor memory throughout late life, whereas participants in a memory failing group exhibited dramatic memory declines late in life. The subgroups were also differentiated by sociodemo- graphic characteristics, functional limitations, and mortality hazards, which may have served as ante- cedents, correlates, or consequents of profile trajectories. We discuss the promises and challenges of using subgroup-oriented approaches in the study of successful aging.
Keywords: successful aging, patterns of aging, old age, mortality, latent profile analysis
Life-span developmentalists are interested in identifying the key components that allow people to lead happy and successful lives (Baltes & Baltes, 1990; Lawton, 1983; Ryff & Singer, 1998). A prominent model advanced by Rowe and Kahn (1997) defined successful aging as a combination of low disease and disability, high levels of cognitive and physical function, and high social engagement. Empirical research testing these notions has primarily
used cross-sectional data to identify subgroups of people aging more or less successfully based on these predefined criteria (e.g., Andrews, Clark, & Luszcz, 2002; Berkman et al., 1993; Garfein & Herzog, 1995; Jorm et al., 1998). However, aging is a process that evolves over time. Conceptual and operational definitions that articulate the dynamic nature of the phenomenon may reveal more nuanced patterns and forms of successful aging. For example,
This article was published Online First March 25, 2013. Jennifer Morack, Department of Human Development and Family Stud-
ies, Pennsylvania State University; Nilam Ram, Department of Human Development and Family Studies, Pennsylvania State University, and Max Planck Institute for Human Development, Berlin, Germany; Elizabeth B. Fauth, Department of Family, Consumer, and Human Development, Utah State University; Denis Gerstorf, Department of Human Development and Family Studies, Pennsylvania State University, and Institute of Psychol- ogy, Humboldt University, Berlin, Germany.
We are grateful for the support provided by the National Institute on Aging (NIA; Grants RC1-AG035645, NIA R21-AG032379, and NIA R21-AG033109), the Max Planck Institute for Human Development, and the Social Science Research Institute at the Pennsylvania State University. We also acknowledge support for the original Swedish studies: National Institutes of Health/NIA Grant R03 AG028471-01, European Union project contract QLK6-CT-2001-02283; Research
Board in the County Council of Jönköping; FORSS; NIA Grant AG- 08861; MacArthur Foundation Research Network on Successful Aging; The Axel and Margaret Axson Johnson’s Foundation; Swedish Council for Social Research; Swedish Foundation for Health Care Sciences and Allergy Research; and NIA Grant T32 AG20500. The content of the article is solely the responsibility of the authors and does not neces- sarily represent the official views of the funding agencies. We also thank Stig Berg, Boo Johansson, Bo Malmberg, Gerald McClearn, Nancy Pedersen, Steven Zarit, and researchers at the Institute for Gerontology in Jönköping University, the Karolinska Institute, and the Pennsylvania State University, who conceived of and completed the original four studies.
Correspondence concerning this article should be addressed to Jennifer Morack, Department of Human Development and Family Studies, Penn- sylvania State University, 422 Biobehavioral Health Building, University Park, PA 16802. E-mail: [email protected]
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
Developmental Psychology © 2013 American Psychological Association 2013, Vol. 49, No. 12, 2309 –2324 0012-1649/13/$12.00 DOI: 10.1037/a0032267
2309
individuals who are functioning well at one point in time may decline considerably afterward and no longer be defined as suc- cessfully aging according to the Rowe and Kahn criteria. Another concern is that using predefined criteria for successful aging may constrain our ability to identify naturally occurring subgroups— individuals who are aging well in distinct aspects of life. Such “uneven” successful aging subgroups, with high functioning in some but not all domains, are repeatedly identified with cross- sectional data (e.g., Garfein & Herzog, 1995; Ko, Berg, Butner, Uchino, & Smith, 2007; Smith & Baltes, 1997). These groups are very informative, in that their existence may provide us with suggestions for how interventions can be targeted for and tailored to specific population segments (e.g., aimed at improving social integration among individuals with overall preserved functioning but a distinct lack of social support).
In this study, we capitalize on the Rowe and Kahn (1997) definition of successful aging with a focus on different aspects of psychosocial functioning, and we go several steps ahead by ex- amining multiple profiles of successful aging as these evolve across old and very old age. In particular, we use a subgroup- oriented approach to understand whether and how we can distin- guish subgroups of persons who show similarities and differences in multidomain trajectories of psychological change. We are in- terested in identifying a set of profiles with uneven patterns of successful (and less successful) aging.
Approaches to the Study of Successful Aging
Research on interrelations among within-person changes has primarily been carried out from a variable-oriented perspective. Here studies describe developmental stability and change in a particular variable and examine how between-person differences in those changes are interrelated with other variables (e.g., how changes in health relate to changes in depressive symptoms). Relations among variables constitute the main focus of analysis, and persons derive their importance from their rank ordering within the overarching distribution of scores within the sample. Such variable-oriented research has provided invaluable insights into normative trajectories of change in a variety of different domains and identified possible mechanisms underlying these changes. For example, a myriad of empirical reports demonstrate that well-being, on average, exhibits relative stability across most of adulthood and old age, with larger decrements only observed with the experience of major social or health-related losses (Ger- storf et al., 2010; Lucas, 2007). In short, well-being appears to be maintained over time and is thus, from this perspective, considered a key component of successful aging.
A person- or subgroup-oriented approach provides a comple- mentary perspective on the study of interrelations among within- person changes (Magnusson, 1998). Here individuals or subgroups of individuals constitute the main focus of analysis, and variables derive their importance from the way they are embedded in the overarching configuration (i.e., profile) of variables within a given person or subgroup. For example, Aldwin, Spiro, Levenson, and Cupertino (2001) used data from the Normative Aging Study to classify men into groups with distinctively different age trajecto- ries in physical and mental health. A subgroup-oriented study may be particularly well suited to empirically test theories of successful aging according to which some groups of people maintain func-
tioning across key domains in old age, whereas other groups of people exhibit steep losses in a variety of different domains (Baltes & Baltes, 1990; Rowe & Kahn, 1997; Ryff & Singer, 1998).
Studies of Successful Aging From a Subgroup- Oriented Perspective
Aside from studies using theoretically or clinically relevant criteria to define groups a priori (e.g., Berkman et al., 1993; Jorm et al., 1998), many subgroup-oriented studies use exploratory approaches such as cluster analysis, latent class, or latent profile analysis (LPA) to identify subgroups within a given sample (e.g., Fiori & Jager, 2012; Gerstorf, Smith, & Baltes, 2006). A compre- hensive overview of subgroup-oriented studies in adult develop- ment and aging is given in the Appendix. It is important to note that although not all of the studies in the Appendix focus on successful aging specifically, each uses an approach that charac- terizes subgroups of individuals across a variety of domains, and the subgroups are often referred to by varying levels of aging well or successfully or are ranked from higher functioning (i.e., suc- cessful aging) to poorer functioning (i.e., less successful aging).
Several cross-sectional studies listed provided highly valuable information about various forms of successful aging. For example, Smith and Baltes (1997) applied cluster analysis to cross-sectional data from older adults (aged 70 –103) in the Berlin Aging Study and identified nine subgroups with distinctively different cross- domain psychological profiles. Although typical findings from variable-oriented studies suggest positive associations between cognitive and social variables (e.g., individuals with poor cogni- tion tend to have low levels of social embeddedness), one of Smith and Baltes’s subgroups was characterized by low cognitive func- tioning and high social embeddedness—an uneven successful ag- ing group counterintuitive to the variable-oriented pattern. That the subgroups were also differentiated by a multitude of variables not used in the group extraction (e.g., health, survival) demonstrated that the profiles provided valid and useful information that would have been missed by a purely variable-oriented perspective.
Although longitudinal subgroup-oriented studies exist (e.g., Lövdén, Bergman, Adolfsson, Lindenberger, & Nilsson, 2005; Maxson, Berg, & McClearn, 1996), very few directly examine the differential aging of subgroups of individuals. Instead, profile configurations at one point in time are often used to predict whether and how those configurations change over time. For example, Gerstorf et al. (2006) applied cluster analysis to data from each of the first three waves of the Berlin Aging Study. Analyses revealed highly similar subgroup profiles over time, and after matching the profiles longitudinally, some two thirds of the participants were found to exhibit stable profiles across time. The groups also showed distinct levels and time-related changes on cognitive, personality, and social domain indicators. Importantly, profile grouping was a robust predictor of long-term outcomes such as mortality, although the subgroup-defining variables were not.
A notable exception of a subgroup-oriented study using longi- tudinal data to examine age changes directly is the report from Aldwin et al. (2001), who first estimated individual growth curves for physical and mental health, and then entered the individual estimates into univariate cluster analyses, one for physical health and one for mental health. Our approach here is an extension of
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2310 MORACK, RAM, FAUTH, AND GERSTORF
this approach by simultaneously including changes occurring in three domains within the profile discovery. Note that this is a somewhat different approach than growth mixture modeling meth- ods (e.g., Maggs & Schulenberg, 2004/2005; Nagin & Tremblay, 1999), which typically define groups based on repeated measures of a univariate outcome. The approach used here explicitly iden- tifies subgroup profiles based on multidimensional change.
The Present Study
We have two major research questions. The first is whether we can identify different types of (un)successful aging using a bottom-up approach and studying naturally occurring subgroups characterized by multivariate trajectory profiles. Specifically, we operationally define successful (psychosocial) aging as sustained high-level functioning in three key domains: maintaining few depressive symptoms, preserving social integration, and maintain- ing good memory. These particular domains were selected because they broadly represent central characteristics of individual func- tioning and psychological development in old age. Depressive symptoms refer to signs of clinically diagnosable depression and provide an important indicator of mental health. Reporting no or very few depressive symptoms over time is one central component of happiness and quality of life (Diener & Seligman, 2002). Social integration revolves around how connected and supported people feel socially, including few feelings of loneliness, and represents a key indicator of the perception of and satisfaction with one’s social life (Cohen, 2004). Cognitive functioning refers to a broad array of mental processes and capacities, with higher functioning consid- ered a general purpose mechanism for adaptation and a resource people draw from to master the challenges of everyday life (Baltes, Lindenberger, & Staudinger, 2006). In particular, memory is the ability of individuals to learn and retain information and is an essential piece of everyday functioning and crucial for indepen- dence later in life.
In our second research question, we corroborate the viability of the profiles by examining how the grouping relates to a set of variables that may have served as antecedents (age, education, gender, marital status, and living arrangement), correlates (change in functional limitations), or consequences (survival time). On the basis of earlier studies (Gerstorf et al., 2006; Ko et al., 2007; Smith & Baltes, 1997), we hypothesize that a sizable group of successful agers indeed maintain psychosocial function throughout late life and that they differ from their less successful peers in key sociode- mographic and health factors. To examine these questions, we applied LPA to 8-year longitudinal data pooled together across four Swedish studies of the oldest old (N � 1,008; Mage � 81 years at Time 1; 61% women) that obtained repeated measures of depressive symptoms, social integration, and memory.
Method
Participants and Procedure
We make use of 8-year longitudinal data pooled across four Swedish longitudinal studies of aging: Sex Differences in Health and Aging study (GENDER; Gold, Malmberg, McClearn, Peder- sen, & Berg, 2002), Swedish Octogenarian study (OCTO; Johans- son & Zarit, 1995), Origins of Variance in the Oldest-Old: Octo-
genarian Twins study (OCTO-TWIN; McClearn et al., 1997), and Swedish Nonagenarian study (NONA; Fauth, Zarit, Malmberg, & Johansson, 2007). In the GENDER and OCTO-TWIN studies, beginning in 1995 and 1990, respectively, representative samples of twin-pairs in their 70s and 80s were recruited from the Swedish Twin Registry, a population-based registry of all multiple births in Sweden. In the OCTO and NONA studies, beginning in 1987 and 1999, respectively, participants in their 80s and 90s were recruited from the municipality of Jönköping’s population registry (which contains names and birth dates of all residents). At baseline as- sessment, OCTO participants were aged 84, 86, 88, and 90 years and NONA participants were aged 86, 90, and 94 years. All four studies followed individuals for three occasions (OCTO-TWIN for five occasions) at 2-year or 4-year (GENDER) intervals.
Our analyses used longitudinal data from the subsample of 1,008 participants who provided data on two or more occasions for the three profile-defining indicators of depressive symptoms, so- cial integration, and memory. These participants were aged be- tween 69 and 95 years at their initial assessment (M � 81.2, SD � 5.6), were 61% women, and provided an average of 3.1 occasions of data over 8 years. Relative to those not included here (N � 785 who did not contribute change information on either of the three profile measures), our participants were younger (M � 81.1, SD � 5.9 vs. M � 84.0, SD � 5.9), F(1, 3526) � 104.54, p � .001; more educated (M � 7.2, SD � 2.3 vs. M � 6.9, SD � 2.1), F(1, 48) � 9.98, p � .01; more likely married (41% vs. 32%), �2(1, N � 1,717) � 16.18, p � .001; and less likely to live in an institution (9% vs. 38%), �2(1, N � 1,774) � 219.88, p � .001; whereas no differences were found for gender. In addition, our subsample reported fewer depressive symptoms (M � 0.5, SD � 0.4 vs. M � 0.7, SD � 0.5), F(1, 10) � 43.32, p � .001; more social integra- tion (M � 3.4, SD � 0.5 vs. M � 3.2, SD � 0.6), F(1, 15) � 49.41, p � .001; greater recall (M � 6.8, SD � 2.5 vs. M � 4.6, SD � 3.4), F(1, 1667) � 200.79, p � .001; and fewer disabilities (M � 1.1, SD � 1.8 vs. M � 4.0, SD � 3.9), F(1, 3274) � 410.99, p � .001. Effect sizes for selectivity differences were in the small to medium range (R2 � .20 for all comparisons).
Relative to participants who provided below the minimum two waves of data for inclusion in our analysis (n � 165), participants who provided three or more waves of data (n � 843) were younger (M � 80.8, SD � 5.7 vs. M � 82.8, SD � 5.5), F(1, 535) � 16.41, p � .001; less likely to live in an institution at Time 1 (7% vs. 19%), �2(1, N � 1,008) � 27.36, p � .001; performed better on the recall test (M � 7.0, SD � 2.5 vs. M � 6.0, SD � 2.7), F(1, 112) � 17.66, p � .001; and had fewer disabilities (M � 0.9, SD � 1.7 vs. M � 1.9, SD � 2.4), F(1, 134) � 41.22, p � .001; but no differences were found for education, marital status, gender, re- porting of depressive symptoms, or social integration. Effect sizes were in the small range (R2 � .04 for all comparisons).
Measures
Profile-defining measures. Measures from three domains representing key areas of psychological functioning (depressive symptoms, social integration, and memory) were used to identify subgroups defined by multidimensional trajectory profiles. Each measure was administered in the same manner in all studies and at each wave, unless otherwise noted. Table 1 provides descriptive information for each measure.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2311PSYCHOLOGICAL PROFILES AND SUCCESSFUL AGING
Depressive symptoms. An individuals’ level of depressive symptoms was indexed by the average of responses to the 10 items from the Center for Epidemiologic Studies Depression Scale (Rad- loff, 1977) that were common across all four pooled studies. Using a scale ranging from 0 (rarely or never) to 3 (most of the time), participants rated how often during the past week they had expe- rienced a variety of depressive symptoms (e.g., thought life had been a failure, felt fearful or depressed; Cronbach’s � � .79).
Social integration. The average of responses to five items measuring subjective support (three items) and loneliness (two items; reverse coded) was used to index social integration (Cron- bach’s � � .74). Participants were asked to rate four of these items (e.g., “Do you have someone you can talk with?”; “Do you feel you are part of a circle of friends?”) adapted from the UCLA Loneliness Scale (Russell, 1982) plus one additional (global) item (Malmberg, 1990) using a scale ranging from 1 (not at all) to 4 (nearly always; for details and measurement properties, see Femia, Zarit, & Johansson, 2001).
Memory. Individuals’ memory was measured with a recall subtest of the Memory in Reality Test (Johansson, 1988/1989). Participants were presented and asked to memorize a list of 10 common objects (keys, medicine, wrist watch, comb, pencil, matchbox, ring, eyeglasses, scissors, and glass). Memory was indexed by the number of words from the list participants recalled when prompted 30 min later (test–retest reliability � .73; for details and measurement properties, see Fiske & Gatz, 2007).
Correlates. We examined whether the groups differed on a variety of factors, including chronological age, years of education, gender (0 � men, 1 � women), marital status (0 � married, 1 � not married), and living arrangement (0 � ordinary housing, 1 � living in an institution). Functional limitations assessed individu- als’ ability to complete four personal activities of daily living (bathing, dressing, toileting, and feeding; Katz, Ford, Moskowitz, Jackson, & Jaffe, 1963) and four instrumental activities of daily living (house cleaning, cooking, shopping, and going places out of walking distance; Lawton, 1971). Participants were asked how much difficulty they had performing those activities from 0 (com- pletely independent) to 3 (unable to do the activity at all). Sum scores of personal activities of daily living and instrumental ac- tivities of daily living were calculated and then averaged to obtain an overall functional limitation score (Fauth, Zarit, & Malmberg,
2008; Cronbach’s � � .82). Finally, survival time was assessed with mortality information obtained from Swedish public health records and quantified as the number of years between an individ- ual’s last assessment and his or her date of death.
Statistical Analysis
In a preliminary step, growth models were used to derive inter- cepts and linear rate of change for depressive symptoms, social integration, and memory that quantified interindividual differences in intraindividual trajectories for the three aspects of psychological functioning. These six measures (level and rate of change for each of the three profile defining variables) were then used in the LPA to identify subgroups of individuals with distinct multidimensional developmental trajectory configurations.
Growth models. Using a multilevel modeling framework, growth curve models (e.g., McArdle & Nesselroade, 2003; Ram & Grimm, 2007; Singer & Willett, 2003) summarized and extracted information about initial levels and rates of change in depressive symptoms, social integration, and memory. Models took the fol- lowing form
Domainti � �0i � �1i�timeti� � eti, (1) where person i’s score in a particular domain at time t, Domainti, is a function of an individual-specific intercept parameter, �0i, and an individual-specific linear slope parameter, �1i, that captures the linear rate of change per year of time, and residual error, eti. Following standard growth curve modeling procedures, individual-specific intercepts, �0i, and linear slopes, �1i, (from the Level 1 model give in Equation 1) were modeled as
�0i � �00 � u0i,
�1i � �10 � u1i, (2)
(i.e., Level 2 model) where �00 and �01 are sample means and u0i and u1i are individual deviations from those means. Using SAS PROC MIXED with restricted maximum likelihood estimation and standard missing at random assumptions (Little & Rubin, 1987), we fitted the model separately for each of the three profile-defining measures. Using Bayes empirical estimates (see Littell, Milliken, Stroup, Wolfinger, & Schabenberger, 2006), we obtained level,
Table 1 Means, Standard Deviations, and Intercorrelations Among the Profile Defining Constructs and Correlates
Construct M SD 1 2 3 4 5 6
Profile defining 1. Depressive symptoms 0.5 0.4 — 2. Social integration 3.4 0.5 �.41� — 3. Memory 6.8 2.5 �.08� .25� —
Correlates 4. Age 81.1 5.8 �.45� .04 �.22� — 5. Years of education 7.2 2.3 .15� �.05 .08� �.14� — 6. Functional limitations 1.1 1.8 �.39� .20� �.31� .37� �.11� — 7. Gender (% women) 61.3 8. Marital status (% married) 41.2 9. Living arrangement (% institutionalized) 8.8
Note. N � 1,008. � p � .05.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2312 MORACK, RAM, FAUTH, AND GERSTORF
�0i, and rate of change, �1i, scores for each domain for each individual. Our objective was to reduce the longitudinal (up to 8 years) data across the three dimensions (depressive symptoms, social integration, memory) down to six informative scores. Given the short length of the individual time series, we did not include a quadratic term in the models.
Latent profile analysis. The estimates of level and rate of change for depressive symptoms, social integration, and memory extracted above were then transformed to Z scores (to alleviate concerns of differential weighting) and used to obtain a set of subgroups defined by their multidimensional trajectory profiles. Specifically, LPA uses latent mixture models to identify latent classes based on mean differences in continuous, manifest vari- ables. We applied LPA to determine the optimal number of sub- groups in the data (Gibson, 1959; Ko et al., 2007; Lanza, Flaherty, & Collins, 2003; Muthén, 2002). Following the standard approach, we estimated profiles and group membership probabilities across a number of alternative models (e.g., group differences in means and/or variances; see Ram & Grimm, 2009). Any number of models with different specifications can be run at the discretion of the researchers and can be based on theory and expectations of the group solution. Using Mplus (Muthén & Muthén, 2006) with maximum likelihood estimation and standard missing at random assumptions (Little & Rubin, 1987), we fitted a series of LPA models with different numbers of classes (two to 10) and possible group differences (means, variances) to the data. Although sub- stantively similar groups were obtained when we allowed both means and variances to differ across groups, we eventually chose not to oversaturate the models with variance differences and rather to prioritize the robustness of results in the face of a sizable amount of sample attrition over time. Thus, only means were allowed to differ across groups. Variances were constrained equal across groups.
In identifying the optimal number of profile groups, we fol- lowed the steps and criteria suggested by Ram and Grimm (2009). First, model parameters were examined for any peculiarities with estimates (e.g., convergence issues, out-of-bounds parameters), and the interpretability of the groupings and estimates were checked. Model solutions that were deemed inappropriate (e.g., convergence problems) were not considered further. Second, re- maining models were compared via model fit criteria including the Bayesian information criteria (BIC) and adjusted BIC. These fit statistics allow for the comparison of models with different num- bers of parameters and penalize for overfitting or having too many parameters in a model. Better fitting models have lower BIC and adjusted BIC (Nylund, Asparouhov, & Muthén, 2007). Third, we considered the entropy statistic for each model, an indicator of the precision of individual profile membership, with values greater than or equal to .80 considered adequate and indicative that indi- viduals are grouped into profiles that described their functional configuration well (Muthén, 2004). Finally, the Vuong–Lo– Mendell–Rubin likelihood ratio test (VLMR-LRT) was used to compare the relative fit of models to similarly structured models with one fewer group. To do so, this test applies a corrected likelihood ratio distribution and provides a p value to test if a model with c groups constitutes a significant improvement in fit relative to a model with c � 1 groups (Lo, Mendell, & Rubin, 2001; Nylund et al., 2007). After determining the best model, subgroup means were transformed back into their raw units.
Group differences. In the final set of analyses, several regression-based methods were used to examine differences among profile groups in relation to sociodemographic and health variables. Specifically, analysis of variance was used to examine group differences in age and education; cross-tabs to examine how the groups differed with respect to gender, marital status, and living arrangement; multilevel models of change to examine group-level differences in longitudinal trajectories of functional limitations; and Cox proportional hazard regression models (Cox, 1972) to test group-level differences in survival hazards.
Results
Subgroup Identification
We used LPA to identify subgroups of individuals differing in their developmental trajectories across depressive symptoms, so- cial integration, and memory. Table 2 shows various fit statistics for models that allowed between two and 10 latent groups. Models with eight and 10 latent groups were immediately eliminated due to nonconvergence, leaving seven models to evaluate with the statistical criteria. With the BIC and adjusted BIC, the nine-group solution offered the best relative fit. Upon further examination, however, many of the groups showed only marginal differences, prompting us to continue examining the other models. For each remaining model, entropy was greater than .80, indicating that all profile solutions had adequate precision of individual profile mem- bership. However, the significant VLMR-LRTs for the two-, three-, and four-group models suggested further consideration of these models—particularly the four-group model. Based on the viability of interpretation and because the BIC and adjusted BIC were smallest and the entropy value (.88) was the highest for the four-class solution, this was selected as the final model. This was further validated by the significant VLMR-LRT for the four-group solution and the nonsignificant one for the five-group solution, indicating a significant improvement in fit from a three- to four- group solution but not from a four- to five-group solution. In sum, our LPA analyses indicated that a four-class model of mean differences provided an optimal description of the data. Table 3 provides an overview of subgroup differences in the six profile-
Table 2 Fit Statistics for Profile Solutions
Profile solution BIC ABIC Entropy VLMR-LRT p value
2 16445 16385 .84 .00 3 16253 16170 .84 .00 4 16077 15972 .88 .00 5 15939 15812 .88 .11 6 15825 15676 .86 .09 7 15742 15571 .87 .62 8a 15667 15474 .88 .09 9 15619 15403 .88 .41
10a 15631 15393 .89 .73
Note. The final model is indicated in bold. BIC � Bayesian information criterion; ABIC � adjusted Bayesian information criterion; VLMR-LRT � Vuong–Lo–Mendell–Rubin likelihood ratio test. a Did not converge.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2313PSYCHOLOGICAL PROFILES AND SUCCESSFUL AGING
defining variables of level and linear rate of change.1 Correspond- ing group-level trajectories of change over time are shown in Figure 1.
Shown in the upper left panel of Figure 1, the largest profile group consisted of 639 individuals (63% of sample). Means for this group were at or above the sample average in all domain indicators and remained so over time. Thus, we labeled the group representing preserved system integrity. Calculated with the sample-level distributions, effect-size metrics indicated that this group’s depressive symptoms were around average at baseline (0.454) and increased at a minor linear rate of change per year (0.005). Social integration and memory were about 0.5 standard deviation above average at baseline (3.505 and 7.586, respectively) and declined at a linear rate that is to be expected for people in this age range (�0.006 and �0.026, respectively).
A second profile group consisted of 176 individuals (18% of sample) and was characterized by relatively low functioning across all domains, particularly in social integration (see upper right panel in Figure 1). This profile was labeled the aging in isolation group. At baseline, this subgroup’s mean scores were about 1 standard deviation above the total sample mean level of depressive symp- toms (0.896), about 1.5 standard deviations below average in social integration (2.492), and 0.5 standard deviation below the sample average in memory (4.522). Over time, this group declined in depressive symptoms about 0.5 standard deviation over the 8 years (linear slope � �0.016), but overall maintained a relatively high level of depressive symptoms. Social integration increased substantially over time (linear slope � 0.073; a little over 1 standard deviation over 8 years), but remained at a low level. Finally, memory performance declined at a to-be-expected, mod- erate rate (linear slope � �0.208).
A third profile group (lower left panel in Figure 1) consisted of 39 individuals (4% of sample) and is referred to as the memory failing group. At baseline, this group’s depressive symptoms were about 0.5 standard deviation lower than the sample average, and their levels of social integration were just around the sample average. Over time, the group maintained their average levels of depressive symptoms, but declined around a 0.5 standard deviation in social integration. The distinguishing feature of this group was the severe loss in memory. This group started off in the average range in memory, but then declined about 3 standard deviations throughout the 8 years of study.
The fourth profile group (lower right panel in Figure 1) con- sisted of 154 individuals (15% of sample) and was labeled the compromised memory group. This group’s trajectories of depres- sive symptoms and social integration were somewhat expected for old and very old individuals and thus considered as normal aging (i.e., around average at baseline and remained fairly stable over time). Of particular interest, memory performance at baseline was around 1.5 standard deviations below the sample average and declined from there at a somewhat average rate (linear slope � �0.318).
After deriving the four-group solution for the whole sample, we ran two additional sets of LPAs to alleviate concerns that the four groups were a by-product of our data configuration. First, we examined the group profiles from only the younger and only the older portions of our sample. The sample was divided in two based on the median age of participants (Group 1 � 72– 84 years; Group 2 � 85–97 years), and we fitted the entire series of models to each subsample. Results for the younger aged sample indicated a three- group solution, with groups resembling the preserved system in- tegrity (n � 405), aging in isolation (but with average recall; n � 59), and compromised memory (n � 40) groups. The older sample was best represented by two groups resembling the preserved system integrity (n � 355) and aging in isolation (n � 149) groups. That three of the four groups emerged (with an expected age selection), and that only the smallest group was missed, provides some level of confidence in the profiles.
Second, we repeatedly split the full sample into random halves and examined the four-group solutions for each half (similar to bootstrapping). In all splits, the preserved system integrity, aging in isolation, and compromised memory profiles were replicated. However, the memory failing profile was always only replicated in
1 Follow-up analyses were performed to determine how each of the four pooled studies was spread out across the subgroups. GENDER participants (who were the youngest of the four studies) were predominantly in the preserved system integrity group (N � 308; 93%). Participants from the remaining studies were split up into the subgroups more evenly, as would be expected from the overall subgroup sizes. Of the OCTO-TWIN partic- ipants, 52% (or n � 239) were in the preserved system integrity and 22% (or n � 239) in the compromised memory group. NONA participants were primarily in the preserved system integrity group (57%, or n � 43), whereas OCTO participants were mainly in the aging in isolation (42%, or n � 61) and preserved system integrity groups (34%, or n � 49).
Table 3 Profiles of the Measures of the Four Subgroups Identified in the Latent Profile Analysis (N � 1,008)
Measure Preserved system
integrity (n � 639; 63%) Aging in isolation (n � 176; 18%)
Memory failing (n � 39; 4%)
Compromised memory (n � 154; 15%)
Overall sample average
Depressive symptoms domain Level 0.454� 0.896� 0.268� 0.341� 0.501�
Rate of change 0.005� �0.016� 0.005 0.005� 0.001 Social domain
Level 3.505� 2.492� 3.376 3.233 3.281�
Rate of change �0.006� 0.073� �0.023� 0.014 0.010�
Memory domain Level 7.586� 4.522� 6.262 1.439� 6.029�
Rate of change �0.026� �0.208� �1.062� �0.318� �0.145�
Note. Measures are in raw units. Rate of change � estimated yearly linear change. � p � .05.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2314 MORACK, RAM, FAUTH, AND GERSTORF
one half of the split, but with a size comparable to what emerged in the whole sample (n � 4% of total). All in all, the format of the replications raised confidence that the four-group solution was viable and may replicate in other samples.
Subgroup Differences in Sociodemographic Characteristics and Health
Next, we explored the utility of the multidimensional profile classifications by examining whether and how these groups dif- fered in a variety of sociodemographic variables (age, education, gender, marital status, living arrangements) and health indicators (functional limitations, survival time).2 One-way analyses of vari- ance indicated significant group differences in age, F(3, 1004) � 82.19, p � .001, and education, F(3, 998) � 5.24, p � .001 (see Figure 2). Scheffé post hoc tests revealed that participants in the preserved system integrity group were, on average, younger (M � 79.2, SD � 5.7) than the other groups and had received signifi- cantly more education (M � 7.4, SD � 2.4) than the compromised memory group (M � 6.8, SD � 1.6), F(3, 998) � 2.61, p � .05.
Chi-square tests of independence revealed group differences in gender, �2(3, N � 1,008) � 11.35, p � .01; marital status, �2 (3, N � 1,008) � 44.28, p � .001; and living arrangement, �2(3, N � 1,008) � 41.88, p � .001. Follow-up comparisons indicated that the preserved system integrity group included fewer women (57%), fewer institutionalized persons (5%), and more married participants (49%) than both the aging in isolation and compro- mised memory groups (see Figure 2). Post hoc analyses of status found that the largest percentage of the aging in isolation group and compromised memory group individuals were widowed (40%
2 For the analyses examining subgroup differences for the variables not used in defining the profiles, a series of simulations were used where individual subgroup membership probabilities were taken into account. One hundred simulated data sets were made, and each was used for every statistical test involving examining subgroup differences for the external variables. Thus, 100 results were obtained for each test. The median test statistic for each analysis was then examined to ensure that it was similar to the result obtained when using the actual sample data. All median statistics were very close to those from the sample data.
Figure 1. The four subgroups identified from latent profile analysis. Each graph represents a different subgroup’s depressive symptoms, social integration, and memory trajectories. The preserved system integrity group maintained average or above-average levels across all domains, whereas the aging in isolation group was relatively low functioning across all domains, particularly in social integration. The memory failing group declined around 3 standard deviations in memory, and the compromised memory group began at around 1.5 standard deviations below average in recall at baseline. Scaling is different for each variable (i.e., means and standard deviations are not the same); therefore, comparisons across domains are not warranted. Also, linear projections extending beyond the range of the scale have been truncated.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2315PSYCHOLOGICAL PROFILES AND SUCCESSFUL AGING
and 45%, respectively), compared to the preserved system integ- rity group and memory failing group individuals who were pri- marily married (49% and 41%, respectively).
A conditional growth model with dummy-coded variables for the groups (with the preserved system integrity group serving as the reference) examined group differences in both initial levels and rates of change for functional limitations. Results revealed signif- icant group differences in average level of functional limitations at baseline, but not in linear rates of change. Figure 3 shows the trajectories of functional limitations for each group. The preserved system integrity and memory failing groups were around 0.5 standard deviation below average at baseline (0.657 and 0.952, respectively, but not statistically distinguishable, p � .47). In comparison, the aging in isolation and compromised memory groups had slightly above-average baseline functional limitations in our sample (2.962 and 2.773, respectively). Over 8 years, rank order differences between groups were maintained.
In our final analysis, Cox (1972) proportional hazard regression models (implemented via the PHREG procedure from the SAS software package (SAS Institute, 1997; see also Allison, 1995) were used to examine group differences in survival hazards, both with and without controlling for age, education, and gender. With the preserved system integrity group as the reference, analyses re- vealed that the relative risks of dying were higher for each of the other three groups: the aging in isolation group, �2(1, N � 1,008) � 28.30, p � .001 (RR � 1.60, CI � [1.34, 1.89]); the memory failing group,
�2(1, N � 1,008) � 14.00, p � .001 (RR � 1.87, CI � [1.35, 2.60]); and the compromised memory group, �2(1, N � 1,008) � 32.32, p � .001 (RR � 1.70, CI � [1.42, 2.04]). The noted group differences remained when controlling for age, gender, and edu- cation. Kaplan–Meier survival curves over the 16-year follow-up period shown in Figure 4 illustrate the group differences in sur-
Figure 2. Differences in subgroups for sociodemographic correlates. Midpoint values on the plots represent rounded averages for each variable. The preserved system integrity group contained the youngest, most educated, least percent women and institutional living, and greatest percent married individuals.
Figure 3. Time-related change in functional limitations for subgroups. Individuals in the preserved system integrity and memory failing group experienced the fewest functional limitations over time.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2316 MORACK, RAM, FAUTH, AND GERSTORF
vival probability. Participants in the preserved system integrity group lived an average of 10 years after baseline assessment, compared to an average of 8 years for the memory failing group, 6 years for the aging in isolation group, and 5 years for the compromised memory group. Taken together, results show the predictive validity of the subgroups over time, with subgroups of individuals who maintain functioning across a variety of domains in late life having better chances for survival.
Discussion
To examine a longitudinal conceptualization of successful ag- ing, we made use of multiple-indicator, multiwave information at the person level and asked directly whether different multivariate configurations of change can be identified. First, we applied LPA to 8-year longitudinal data from 1,008 oldest-old adults to distin- guish four multidimensional profiles of change that capture dis- tinct differences in trajectories of depressive symptoms, social integration, and memory: a preserved system integrity group of participants who maintained functioning across very old age; an aging in isolation group with a persistent lack of social support; and two groups of people with average well-being and social functions but distinctive memory profiles: A compromised mem- ory group was characterized by poor memory throughout late life, whereas participants in a memory failing group experienced dra- matic memory declines late in life. Second, we corroborated the distinctiveness of the profiles by demonstrating group differences in a variety of sociodemographic and health indicators that were not involved in defining the profiles. We discuss how the longi- tudinal multivariate profiles reflect various configurations of suc- cessful and less successful aging very late in life. In doing so, we consider how a subgroup-oriented approach offers both possibili- ties and challenges for obtaining better understanding of the intri- cacies of human development.
Subgroups
Preserved system integrity group. The largest profile group consisting of 639 participants (63% of sample) was uniquely
characterized by average to above-average trajectories across the psychological domains examined. The label preserved system in- tegrity denoted that the three systems of functioning changed together in a way that sustained effective psychosocial functioning over time. In line with this interpretation, participants in this group experienced better physical functioning, displaying fewer func- tional limitations and living, on average, longer than individuals in other groups. Of note is that participants in this profile were relatively younger, more educated, and more likely to be a man, married, and live on their own, each of which may be considered protective against late-life declines. For example, having close social relationships, especially marital partners, has been linked to better health outcomes and a longer life (Seeman & Crimmins, 2001).
The high levels of functioning in the preserved system integrity group illustrate the concept of successful aging and mirror the high-functioning profile groups reported in several other subgroup-oriented studies, such as the high-cognitive, high-social, high-well-being group in Maxson et al. (1996); the general positive profile in Smith and Baltes (1997); the overall positive profile in Gerstorf et al. (2006); and the generally positive group in Ko et al. (2007). We note that— consistent across studies, including our own report—the majority of participants were grouped into the profiles with across-domain high levels of functioning. For exam- ple, Gerstorf et al. and Ko et al. reported that some 50% of their samples were categorized as aging successfully. As an extension of these reports, we defined our groups based on cross-sectional differences at Time 1 when the average age of the sample was 81 years as well as based on up to 8 years of longitudinal change in key domains of psychological functioning—a time frame that should be long enough to exhibit change in this age range. We thus found it striking that 63% of oldest-old participants were classified into such a successful aging group who maintained their function- ality over time. Of course, this finding can be taken to exemplify the positively select nature of our sample of very old adults and questions whether the finding extends to the overall population. At the same time, we note that it was only through the use of a subgroup-oriented approach that we were able to identify directly the existence of such a successful-aging group and the persistence of systemic integrity across very old age. It will be instructive to explore whether bottom-up approaches are better geared toward identifying successfully aging oldest-old than using strict a priori criteria in a top-down manner (e.g., review by Depp & Jeste, 2006: 36% of participants identified as aging successfully).
Aging in isolation group. Compared to the preserved system integrity group, the aging in isolation group (N � 176, 18% of sample) exhibited poor functioning in all three domains. The label aging in isolation was selected to highlight the distinctive lack of social integration even in the context of poor functioning in other domains. That is, despite needing support, this group did not appear to be embedded in social networks that might provide that support. Adding to the overall picture, participants in this group were, compared to the other profiles, the most likely to not be married (anymore) and to live in institutions, and participants also had the highest level of functional limitations. Again, these char- acteristics very closely match some of the profiles reported in earlier subgroup-oriented studies, such as the overall negative group in Maxson et al. (1996); the general negative profile in Smith and Baltes (1997); and the extremely frail, lonely, depressed
Figure 4. Differences between subgroups in survival probabilities over 17 years. The preserved system integrity group lived longer on average than the other profiles.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2317PSYCHOLOGICAL PROFILES AND SUCCESSFUL AGING
group in Smith and Baltes (1998). Our study adds another piece to the puzzle by identifying social isolation as a distinctive charac- teristic of poor functioning profile groups.
According to the criteria suggested by Rowe and Kahn (1997), the aging in isolation profile could be considered the least suc- cessful agers of the four groups identified here. For this group, functioning was not preserved in any of the three domains of health (high functional limitations), cognitive abilities (poor mem- ory), and active engagement with life (low levels of social inte- gration). Importantly, we note that this group included only 18% of our sample, illustrating that poor functioning is not a necessary characteristic of advanced old age.
Although difficult to examine and not explored in this study, it is possible that loneliness and a lack of social support may have con- tributed to poor functioning in the other domains (Berkman, Glass, Brissette, & Seeman, 2000). In particular, research by Cacioppo, Hawkley, Norman, and Bernsten (2011) suggests that social isolation is related to elevated hazards for morbidity and mortality in older adults through multiple biological pathways. The identification of this group suggests that for a portion of older adults, social integration may be a vital component of systemic function— one that provides a potential lever for intervention. Participation in productive activities and social support may buffer declines in other domains, whereas lack thereof may burden overall functioning. However, with the data at hand, we cannot draw any temporal inferences. As a consequence, it is possible that depressive symptoms may have been a source for reporting small social networks and feeling socially isolated (which may or may not coincide with the objectively available support). Of note is that from a variable-oriented perspective, the two constructs of depressive symptoms and social integration showed only moderate overlap (r � �.41 in the entire sample; r � �.35 in the aging in isolation group).
Memory failing group. Participants in the memory failing group (N � 39, 4% of sample) were characterized by average levels of depressive symptoms and social integration throughout the course of the study, combined with a dramatic decline in memory. Such severe decrements in memory abilities may indicate that participants were beginning to experience some form of de- mentia (testing this was beyond the scope of the current analyses). Interestingly, the memory failing group was comparable to the preserved system integrity group on all external variables (age, education, etc.). However, we do not rule out the possibility that marker variables could be identified in larger samples. Our follow-up analyses indicated that although the profile seems to exist, the small subgroup size restricted our ability to extract it in all subsamples. It is instrumental for future research to examine the existence of such a group in more detail.
The precipitous memory declines distinctive of the memory failing group match profiles reported by Lövdén, Bergman, et al. (2005) from the Betula study. In particular, the authors showed that particular groups of participants experienced a developmental cascade of steep declines in several cognitive abilities (including episodic and semantic memory) that eventually culminated in dementia diagnosis and death. In both our study and the Betula project, the size of the dementia-prone subgroup was relatively small, suggesting that this is an expected scenario for only a small portion of study samples, but not the large majority of participants typically included in large macro-longitudinal studies. To the extent that findings from our positively select subsample of Swed-
ish oldest old can be generalized to the larger population, it appears that severe forms of memory decline may be less common than often stereotypically expected. Our analysis of multidimensional change profiles allowed for the identification of an uneven profile of successful aging, with distinct declines in one domain in the context of preserved functioning in other domains. If information from Time 1 only had been used in the profile-defining stage, participants in this group may have been lumped into the success- ful aging profile because of above-average memory at baseline. In our view, this provides an excellent example of the importance of using longitudinal data to characterize successful aging groups.
Compromised memory group. The compromised memory group (N � 154, 15% of sample) maintained fairly average levels of depressive symptoms and social integration, but was uniquely char- acterized by very poor memory ability throughout the course of the study. Relative to the memory failing group, this group did not decline dramatically in memory. Instead, those in this group already had low ability levels at the start of observation and remained low over time. They were also older than the rest of the sample and had the highest mortality hazards. The compromised memory group resembles groups reported from other subgroup-oriented studies, such as the cognitively impaired and high external control group in Smith and Baltes (1997) and the low-cognitive-functioning groups reported from Ko et al. (2007) and Maxson et al. (1996).
Given that the profile-defining domains typically show a posi- tive manifold, a disparate profile like the compromised memory group provides another illustrative example of possible insights gained by pursuing a subgroup-oriented approach that would have been missed by a variable-oriented analysis. It remains an open question, though, why (poor) levels of memory were maintained over time. As suggested by more variable-oriented research (Ger- storf, Lövdén, Röcke, Smith, & Lindenberger, 2007; Lövdén, Ghisletta, & Lindenberger, 2005), it is possible that the relatively stable and average levels of depressive symptoms and social integration have protected against memory decline. The compro- mised memory group provides another example of an uneven profile, with distinctively low functioning in one domain (mem- ory) and fairly typical functioning in other domains. Among the questions to be addressed in future research is whether and how uneven changes qualify as (un)successful aging.
Potentials and Challenges of a Subgroup-Oriented Approach
Adopting a subgroup-oriented approach with multiple domains, our study sheds additional light on questions about differential and successful development in old age (e.g., Baltes, 1987; Birren, 1959; Dannefer, 2003; Riley, 1987). We identified multidimen- sional configurations of psychological change that would be dif- ficult to obtain with a variable-oriented approach. For example, two subgroups of individuals showed average functioning in de- pressive symptoms and social integration but distinctively differ- ent trajectories of memory change (memory failing and compro- mised memory groups). We provide direct evidence that associations found at the sample level only apply to some sub- groups of individuals but not to others—and thus continue to highlight the risks of using between-person associations to sub- stantiate within-person theory (Estes, 1956). Of course, we have only gone partially in this direction. The models still rest on assump-
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2318 MORACK, RAM, FAUTH, AND GERSTORF
tions of within-group homogeneity, aggregate across individuals, and probabilistically assign individuals to subgroup-level profiles that are considered representative of individual-level profiles (rather than uti- lizing probabilities of membership in all four profiles).
As alternatives to the variable-oriented stance gain momentum, our study builds on the other subgroup-oriented studies of old age listed in the Appendix. In particular, our extension makes use of data on within-person changes across multiple domains of psychological function to identify multidimensional-longitudinal, and arguably more holistic, profiles. The earlier subgroup-oriented research rarely mod- eled within-person change directly. Rather, they identified groups from cross-sectional data and then used the grouping variable to predict subsequent change. Addressing this gap in the application of latent class methods, we identified profiles directly from the longitu- dinal change scores. Also, in contrast to some group-based trajectory models (Nagin & Odgers, 2010), that all individuals within a sub- group follow the same developmental trajectory, we chose a statistical approach that allowed for within-subgroup heterogeneity (see Ram, Grimm, Gatzke-Kopp, & Molenaar, 2012, for discussion of the po- tential hazards of both approaches).
The relative cost/utility of variable-oriented and subgroup-oriented approaches must be considered. Although both types of analysis can be done in the flash of an eye, the time-consuming and effortful nature of multivariate long-term panel data must also be considered. Cer- tainly, a multivariate analysis makes better use of the data than a univariate analysis, but we ourselves still struggle with the added utility of a purely subgroup-oriented approach. Case in point, we interpreted and named the groups from a variable-oriented perspec- tive, and following standard good practices, we evaluated the utility of the profile groups by assessing how the categorical grouping variable was related to a variety of antecedents, correlates, and outcomes in a purely variable-oriented manner. In sum, the field still struggles to take a subgroup-oriented approach without placing it within variable- oriented interpretations and analysis.
A practical benefit of subgroup-oriented interpretation emerges when considering potential interventions. Our results point toward tailoring diagnostic and intervention efforts to individual needs. From a clinical perspective, for example, assessments aimed at risk profiles are often based on cross-sectional data or examine risk factors in a univariate manner. Our findings suggest that profiles based on mul- tidimensional, longitudinal data can highlight the specific risk and protective factors that may be vital for achieving successful aging outcomes, including quality of life, a delayed entry into institutional- ization, and lower mortality hazards. For example, the aging in iso- lation group may require interventions that aim to promote social support, alleviate depressive symptoms, and enhance cognitive re- serve, whereas the compromised memory or memory failing groups may benefit from interventions fully devoted to enhancing cognitive reserve or delaying future cognitive loss.
Limitations and Outlook
To broadly represent central characteristics of successful psy- chosocial functioning, we selected three domains and well- established indicators thereof from measures available in the four Swedish studies of old age. Of course, the inclusion of both additional domains (e.g., activities, motivation, personality, brain efficiency) and additional indicators per domain (e.g., measures of crystallized cognitive abilities) would likely paint a more refined
profile landscape. In particular, the measure of social integration did not capture all aspects of what it means to be socially inte- grated. Social integration is not just a lack of loneliness and feeling a part of a group, but also being involved in an array of activities and relationships with significant others (Cohen, 2004). As well, we made use of linear change scores derived from biyearly re- peated measures. Shorter time intervals and more frequent obser- vations would allow for more refined assessments of within-person change to put into the profile analyses. Particularly informative may be change measures that quantify differences in the timing, tempo, and asymmetry of developmental changes (see Grimm & Ram, 2009; Visser, 2011). In this context, we also note that our analyses of time-varying predictors (e.g., functional limitations) did not allow for disentangling whether these served as precursors or consequences of psychosocial profiles. We can only infer that changes in psychosocial profiles were accompanied by changes in functional health. In addition to the constraints invoked by the available repeated measures, results found in our positively select subpopulation of the oldest old in Sweden may not necessarily generalize to other population segments. For example, participants in the sample consistently reported very low levels of depressive symptoms, which may not well characterize certain subpopulations and/or the diversity of nations such as the United States where tremendous differences prevail in socioeconomic, ethnic, and re- source backgrounds. Finally, our pooling method did not allow for description of how subgroups emerge and change over late adult- hood. From the follow-up LPAs conducted on the younger and older portions of the sample, it is possible that the number of subgroups decreases with age, and this should be explored in more detail in future work.
In closing, we note that our study departed from the Rowe and Kahn (1997) model in several important ways. First, we defined successful aging based entirely on psychosocial domains of func- tioning. It was only in a second step that other key domains such as physical health were considered. Second, rather than operation- ally defining successful aging in a cross-sectional manner, we directly modeled multiwave longitudinal data in three key domains to identify groups of people who age more or less successfully. Making use of a subgroup-oriented approach, our results revealed four successful and less successful aging profiles with distinct trajectories of change across indicators of depressive symptoms, social integration, and memory. These profiles were also differen- tiated by sociodemographic characteristics, functional limitations, and mortality hazards that may have served as antecedents, corre- lates, or consequences of the profile trajectories. We take these insights to illustrate the utility of using multiple-indicator infor- mation at the person level in advancing our understanding of the differential aging and successful aging of individuals. Results also provide impetus to further address the promises and challenges of a subgroup-oriented approach.
References
Aldwin, C. M., Spiro, A., Levenson, M. R., & Cupertino, A. P. (2001). Longitudinal findings from the Normative Aging Study: III. Personality, individual health trajectories, and mortality. Psychology and Aging, 16, 450 – 465. doi:10.1037/0882-7974.16.3.450
Allison, P. D. (1995). Survival analysis using the SAS system: A practical guide. Cary, NC: SAS Institute.
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2319PSYCHOLOGICAL PROFILES AND SUCCESSFUL AGING
Andrews, G., Clark, M., & Luszcz, M. (2002). Successful aging in the Australian Longitudinal Study of Aging: Applying the MacArthur model cross-nationally. Journal of Social Issues, 58, 749 –765. doi:10.1111/ 1540-4560.00288
Baltes, P. B. (1987). Theoretical propositions of life-span developmental psychology: On the dynamics between growth and decline. Develop- mental Psychology, 23, 611– 626. doi:10.1037/0012-1649.23.5.611
Baltes, P. B., & Baltes, M. M. (Eds.). (1990). Successful aging: Perspec- tives from the behavioral sciences. New York, NY: Cambridge Univer- sity Press.
Baltes, P. B., Lindenberger, U., & Staudinger, U. M. (2006). Life-span theory in developmental psychology. In R. M. Lerner & W. Damon (Eds.), Handbook of child psychology: Vol. 1. Theoretical models of human development (6th ed., pp. 569 – 664). New York, NY: Wiley.
Berkman, L. F., Glass, T., Brissette, I., & Seeman, T. E. (2000). From social integration to health: Durkheim in the new millennium. Social Science & Medicine, 51, 843– 857. doi:10.1016/S0277-9536(00)00065-4
Berkman, L. F., Seeman, T. E., Albert, M., Blazer, D., Kahn, R., Mohs, R., . . . Rowe, J. (1993). High, usual and impaired functioning in community-dwelling older men and women: Findings from the MacAr- thur Foundation Research Network on Successful Aging. Journal of Clinical Epidemiology, 46, 1129 –1140. doi:10.1016/0895- 4356(93)90112-E
Birren, J. E. (1959). Principles of research on aging. In J. E. Birren (Ed.), Handbook of aging and the individual: Psychological and biological aspects (pp. 3– 42). Chicago, IL: University of Chicago Press.
Bosworth, H. B., & Schaie, K. W. (1997). The relationship of social environment, social networks, and health outcomes in the Seattle Lon- gitudinal Study: Two analytical approaches. Journals of Gerontology: Series B. Psychological Sciences and Social Sciences, 52, P197–P205. doi:10.1093/geronb/52B.5.P197
Cacioppo, J. T., Hawkley, L. C., Norman, G. J., & Bernsten, G. C. (2011). Social isolation. Annals of the New York Academy of Sciences, 1231, 17–22. doi:10.1111/j.1749-6632.2011.06028.x
Cohen, S. (2004). Social relationships and health. American Psychologist, 59, 676 – 684. doi:10.1037/0003-066X.59.8.676
Cox, D. R. (1972). Regression models and life-tables. Journal of the Royal Statistical Society: Series B. Methodological, 34, 187–220.
Dannefer, D. (2003). Cumulative advantage/disadvantage and the life course: Cross-fertilizing age and social science theory. Journals of Gerontology: Series B. Psychological and Social Sciences, 58, S327– S337. doi:10.1093/geronb/58.6.S327
Depp, C. A., & Jeste, D. V. (2006). Definitions and predictors of successful aging: A comprehensive review of larger quantitative studies. American Journal of Geriatric Psychiatry, 14, 6 –20. doi:10.1097/01.JGP .0000192501.03069.bc
Diener, E., & Seligman, M. E. (2002). Very happy people. Psychological Science, 13, 81– 84. doi:10.1111/1467-9280.00415
Estes, W. K. (1956). The problem of inference from curves based on group data. Psychological Bulletin, 53, 134 –140. doi:10.1037/h0045156
Evert, J., Lawler, E., Bogan, H., & Perls, T. (2003). Morbidity profiles of centenarians: Survivors, delayers, and escapers. Journals of Gerontol- ogy: Series A. Biological Sciences and Medical Sciences, 58, M232– M237. doi:10.1093/gerona/58.3.M232
Fauth, E. B., Zarit, S. H., & Malmberg, B. (2008). Mediating relationships within the disablement process model: A cross-sectional study of the oldest-old. European Journal of Ageing, 5, 161–179. doi:10.1007/ s10433-008-0092-6
Fauth, E. B., Zarit, S. H., Malmberg, B., & Johansson, B. (2007). Physical, cognitive, and psychosocial variables from the disablement process model predict patterns of change in disability for the oldest-old. Geron- tologist, 47, 613– 624. doi:10.1093/geront/47.5.613
Femia, E. E., Zarit, S. H., & Johansson, B. (2001). The disablement process in very late life: A study of the oldest-old in Sweden. Journals of
Gerontology: Series B. Psychological Sciences and Social Sciences, 56, P12–P23. doi:10.1093/geronb/56.1.P12
Fiori, K. L., Antonucci, T. C., & Cortina, K. S. (2006). Social network typologies and mental health among older adults. Journals of Gerontol- ogy: Series B. Psychological Sciences and Social Sciences, 61, P25–P32. doi:10.1093/geronb/61.1.P25
Fiori, K. L., & Jager, J. (2012). Social support networks and health across the lifespan: A longitudinal, pattern-centered approach. International Journal of Behavioral Development, 36, 117–129. doi:10.1177/ 0165025411424089
Fiori, K. L., Smith, J., & Antonucci, T. C. (2007). Social network types among older adults: A multidimensional approach. Journals of Geron- tology: Series B. Psychological Sciences and Social Sciences, 62, P322– P330. doi:10.1093/geronb/62.6.P322
Fiske, A., & Gatz, M. (2007). The Apartment Test: Validity of a memory measure. Aging, Neuropsychology, and Cognition, 14, 441– 461. doi: 10.1080/13825580600611294
Garfein, A. J., & Herzog, A. R. (1995). Robust aging among the young-old, old-old, and oldest-old. Journals of Gerontology: Series B. Psycholog- ical Sciences and Social Sciences, 50, S77–S87. doi:10.1093/geronb/ 50B.2.S77
Gerstorf, D., Lövdén, M., Röcke, C., Smith, J., & Lindenberger, U. (2007). Well-being affects changes in perceptual speed in advanced old age: Longitudinal evidence for a dynamic link. Developmental Psychology, 43, 705–718. doi:10.1037/0012-1649.43.3.705
Gerstorf, D., Ram, N., Mayraz, G., Hidajat, M., Lindenberger, U., Wagner, G. G., & Schupp, J. (2010). Late-life decline in well-being across adulthood in Germany, the United Kingdom, and the United States: Something is seriously wrong at the end of life. Psychology and Aging, 25, 477– 485. doi:10.1037/a0017543
Gerstorf, D., Smith, J., & Baltes, P. B. (2006). A systemic–wholistic approach to differential aging: Longitudinal findings from the Berlin Aging Study. Psychology and Aging, 21, 645– 663. doi:10.1037/0882- 7974.21.4.645
Gibson, W. A. (1959). Three multivariate models: Factor analysis, latent structure analysis, and latent profile analysis. Psychometrika, 24, 229 – 252. doi:10.1007/BF02289845
Gold, C. H., Malmberg, B., McClearn, G. E., Pedersen, N. L., & Berg, S. (2002). Gender and health: A study of unlike gender twins. Journals of Gerontology: Series B. Psychological Sciences and Social Sciences, 57, S168 –S176. doi:10.1093/geronb/57.3.S168
Grimm, K. J., & Ram, N. (2009). Nonlinear growth models in Mplus and SAS. Structural Equation Modeling, 16, 676 –701. doi:10.1080/ 10705510903206055
Hsu, H.-C., & Jones, B. L. (2012). Multiple trajectories of successful aging of older and younger cohorts. The Gerontologist, 52, 843– 856. doi: 10.1093/geront/gns005
Johansson, B. (1988/1989). The MIR–Memory in Reality Test. Stockholm, Sweden: Psykologiförlaget.
Johansson, B., & Zarit, S. H. (1995). Prevalence and incidence of dementia in the oldest-old: A longitudinal study of a population-based sample of 84 –90-year-olds in Sweden. International Journal of Geriatric Psychi- atry, 10, 359 –366. doi:10.1002/gps.930100504
Jorm, A. F., Christensen, H., Henderson, A. S., Jacomb, P. A., Korten, A. E., & Mackinnon, A. (1998). Factors associated with successful ageing. Australasian Journal on Ageing, 17, 33–37. doi:10.1111/j.1741- 6612.1998.tb00222.x
Katz, S., Ford, A. B., Moskowitz, R. W., Jackson, B. A., & Jaffe, M. W. (1963). Studies of illness and the aged: The index of ADL: A standard- ized measure of biological and psychosocial function. Journal of the American Medical Association, 185, 914 –923. doi:10.1001/jama.1963 .03060120024016
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2320 MORACK, RAM, FAUTH, AND GERSTORF
Ko, K. J., Berg, C. A., Butner, J., Uchino, B. N., & Smith, T. W. (2007). Profiles of successful aging in middle-aged and older married couples. Psychology and Aging, 22, 705–718. doi:10.1037/0882-7974.22.4.705
Lanza, S., Flaherty, B. P., & Collins, L. M. (2003). Latent class and latent transition analysis. In J. A. Schinka & W. F. Velicer (Eds.), Handbook of psychology: Research methods in psychology (Vol. 2, pp. 663– 685). New York, NY: Wiley.
Lawton, M. P. (1971). The functional assessment of older people. Journal of the American Geriatric Society, 19, 465– 481.
Lawton, M. P. (1983). Environment and other determinants of well-being in older people. The Gerontologist, 23, 349 –357. doi:10.1093/geront/23 .4.349
Littell, R. C., Milliken, G. A., Stroup, W. W., Wolfinger, R. D., & Schabenberger, O. (2006). SAS for mixed models (2nd ed.). Cary, NC: SAS Institute.
Little, R. J. A., & Rubin, D. B. (1987). Statistical analysis with missing data. New York, NY: Wiley.
Litwin, H. (1995). The social networks of elderly immigrants: An analytic typology. Journal of Aging Studies, 9, 155–174. doi:10.1016/0890- 4065(95)90009-8
Litwin, H. (2001). Social network type and morale in old age. The Ger- ontologist, 41, 516 –524. doi:10.1093/geront/41.4.516
Lo, Y., Mendell, N. R., & Rubin, D. B. (2001). Testing the number of components in a normal mixture. Biometrika, 88, 767–778. doi:10.1093/ biomet/88.3.767
Lövdén, M., Bergman, L., Adolfsson, R., Lindenberger, U., & Nilsson, L.-G. (2005). Studying individual aging in an interindividual context: Typical paths of age-related, dementia-related, and mortality-related cognitive development in old age. Psychology and Aging, 20, 303–316. doi:10.1037/0882-7974.20.2.303
Lövdén, M., Ghisletta, P., & Lindenberger, U. (2005). Social participation attenuates decline in perceptual speed in old and very old age. Psychol- ogy and Aging, 20, 423– 434. doi:10.1037/0882-7974.20.3.423
Lucas, R. E. (2007). Adaptation and the set-point model of subjective well-being: Does happiness change after major life events? Current Directions in Psychological Science, 16, 75–79. doi:10.1111/j.1467- 8721.2007.00479.x
Maggs, J. L., & Schulenberg, J. (2004/2005). Trajectories of alcohol use during the transition to adulthood. Alcohol Research & Health, 28, 195–201.
Magnusson, D. (1998). The logic and implication of a person approach. In R. B. Cairns, L. R. Bergman, & J. Kagan (Eds.), Methods and models for studying the individual (pp. 33– 63). Thousand Oaks, CA: Sage.
Malmberg, B. (1990). Access to resources in different age-cohorts: Impli- cations for activity level, loneliness and life satisfaction (Unpublished doctoral dissertation). Linköping University, Linköping, Sweden.
Manton, K. G., & Land, K. C. (2000). Multidimensional disability/ mortality trajectories at ages 65 and over: The impact of state depen- dence. Social Indicators Research, 51, 193–221. doi:10.1023/A: 1006966525964
Manton, K. G., Siegler, I. C., & Woodbury, M. A. (1986). Patterns of intellectual development in later life. Journal of Gerontology, 41, 486 – 499. doi:10.1093/geronj/41.4.486
Maxson, P. J., Berg, S., & McClearn, G. (1996). Multi-dimensional pat- terns of aging in 70-year-olds: Survival differences. Journal of Aging and Health, 8, 320 –333. doi:10.1177/089826439600800302
McArdle, J. J., & Nesselroade, J. R. (2003). Growth curve analysis in contemporary psychological research. In J. Schinka & W. Velicer (Eds.), Comprehensive handbook of psychology: Volume 2. Research methods in psychology (pp. 447– 480). New York, NY: Pergamon Press.
McClearn, G. E., Johansson, B., Berg, S., Pedersen, N. L., Ahern, F., Petrill, S. A., & Plomin, R. (1997). Substantial genetic influence on cognitive abilities in twins 80 or more years old. Science, 276, 1560 – 1563. doi:10.1126/science.276.5318.1560
Muthén, B. (2002). Beyond SEM: General latent variable modeling. Be- haviormetrika, 29, 81–117. doi:10.2333/bhmk.29.81
Muthén, B. (2004). Latent variable analysis. Growth mixture modeling and related techniques for longitudinal data. In D. Kaplan (Ed.), The Sage handbook of quantitative methodology for the social sciences (pp. 345– 368). Thousand Oaks, CA: Sage.
Muthén, L. K., & Muthén, B. O. (2006). Mplus user’s guide (4th ed.). Los Angeles, CA: Muthén & Muthén.
Nagin, D. S., & Odgers, C. L. (2010). Group-based trajectory modeling in clinical research. Annual Review of Clinical Psychology, 6, 109 –138. doi:10.1146/annurev.clinpsy.121208.131413
Nagin, D., & Tremblay, R. E. (1999). Trajectories of boys’ physical aggression, opposition, and hyperactivity on the path to physically violent and nonviolent juvenile delinquency. Child Development, 70, 1181–1196. doi:10.1111/1467-8624.00086
Neugarten, B. L., Havighurst, R. J., & Tobin, S. S. (1968). Personality and patterns of aging. In B. L. Neugarten (Ed.), Middle age and aging (pp. 173–177). Chicago, IL: University of Chicago Press.
Nylund, K. L., Asparouhov, T., & Muthén, B. (2007). Deciding on the number of classes in latent class analysis and growth mixture modeling: A Monte Carlo simulation study. Structural Equation Modeling, 14, 535–569. doi:10.1080/10705510701575396
Radloff, L. S. (1977). The CES-D Scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385– 401. doi:10.1177/014662167700100306
Ram, N., & Grimm, K. J. (2007). Using simple and complex growth models to articulate developmental change: Matching method to theory. International Journal of Behavioral Development, 31, 303–316. doi:10 .1177/0165025407077751
Ram, N., & Grimm, K. J. (2009). Growth mixture modeling: A method for identifying differences in longitudinal change among unobserved groups. International Journal of Behavioral Development, 33, 565–576. doi:10.1177/0165025409343765
Ram, N., Grimm, K. J., Gatzke-Kopp, L. M., & Molenaar, P. C. M. (2012). Longitudinal mixture models and the identification of archetypes. In B. Laursen, T. D. Little, & N. A. Card (Eds.) Handbook of developmental research methods (pp. 481–500). New York, NY: Guilford Press.
Riley, M. W. (1987). Aging, health, and social change: An overview. In M. W. Riley & J. D. Matarazzo (Eds.), The aging dimension: Perspec- tives in behavioral medicine (pp. 1–14). Hillsdale, NJ: Erlbaum.
Rowe, J. W., & Kahn, R. L. (1997). Successful aging. The Gerontologist, 37, 433– 440. doi:10.1093/geront/37.4.433
Russell, D. (1982). The measurement of loneliness. In L. A. Peplau & D. Perlman (Eds.), Loneliness: A sourcebook of current theory, research and therapy (pp. 81–104). New York, NY: Wiley.
Ryff, C. D., & Singer, B. (1998). The contours of positive human health. Psychological Inquiry, 9, 1–28. doi:10.1207/s15327965pli0901_1
SAS Institute. (1997). SAS/STAT software: Changes and enhancements through Release 6.12. San Diego, CA: Academic Press.
Seeman, T. E., & Crimmins, E. (2001). Social environment effects on health and aging. Annals of the New York Academy of Sciences, 954, 88 –117. doi:10.1111/j.1749-6632.2001.tb02749.x
Singer, B., Ryff, C. D., Carr, D., & Magee, W. J. (1998). Linking life histories and mental health: A person-centered strategy. Sociological Methodology, 28, 1–51. doi:10.1111/0081-1750.00041
Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data analysis: Methods for studying change and event occurrence. New York, NY: Oxford University Press.
Smith, J., & Baltes, M. M. (1998). The role of gender in very old age: Profiles of functioning and everyday life patterns. Psychology and Aging, 13, 676 – 695. doi:10.1037/0882-7974.13.4.676
Smith, J., & Baltes, P. B. (1997). Profiles of psychological functioning in the old and oldest old. Psychology and Aging, 12, 458 – 472. doi: 10.1037/0882-7974.12.3.458
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2321PSYCHOLOGICAL PROFILES AND SUCCESSFUL AGING
Visser, I. (2011). Seven things to remember about hidden Markov models: A tutorial on Markovian models for time series. Journal of Mathematical Psychology, 55, 403– 415. doi:10.1016/j.jmp.2011.08.002
Wickrama, K. A. S., Mancini, J. A., Kwag, K., & Kwon, J. (2012). Heterogeneity in multidimensional health trajectories of late old years and socioeconomic stratification: A latent trajectory class analysis. Jour- nals of Gerontology: Series B. Psychological Sciences and Social Sci- ences, 68, 290 –297. doi:10.1093/geronb/gbs111
Wong, J. D., & Hardy, M. A. (2009). Women’s retirement expectations: How stable are they? Journals of Gerontology: Series B. Psycholog- ical Sciences and Social Sciences, 64, 77– 86. doi:10.1093/geronb/ gbn010
Zarit, S. H., Johansson, B., & Berg, S. (1993). Functional impairment and co-disability in the oldest old: A multidimensional approach. Journal of Aging and Health, 5, 291–305. doi:10.1177/089826439300500301
Appendix
Examples of Studies Applying a Subgroup- or Subgroup-Oriented Approach
Study Wave N Group defining domains External variables Grouping method Group findings
Cross-sectional studies Kansas City Study of
Adult Life (Neugarten et al., 1968)
1 59 Activity, personality, well- being
Predefined top-down procedure
Eight groups characterized primarily by personality and also activity and well- being
MacArthur Studies on Successful Aging (Berkman et al., 1993)
1 1,354 Cognition, functional health Psychosocial, physiological, sociodemographic
Predefined by successful aging criteria
Three groups: high, middle, and low functioning
Swedish Octogenarian study (Zarit et al., 1993)
1 320 Cognition, functional health, sensory functioning
Examination of disability base rates and degree of codisability
Three groups: no significant impairments, impairments in instrumental activities of daily living only, and codisability
Americans’ Changing Lives Survey (Garfein & Herzog, 1995)
1 1,644 Cognition, functional health, productivity, well-being
Health, personality, psychosocial, sociodemographic
Predefined robust aging criteria for each domain
Groups identified for each domain (four groups for each domain except functional health with five groups)
Data from study on Russian Jewish immigrants in Israel (Litwin, 1995)
1 259 Social network Social support, sociodemographic
Cluster analysis Four social network types
Seattle Longitudinal Study (Bosworth & Schaie, 1997)
1 387 Social integration Health, sociodemographic
Cluster analysis Four groups characterized by levels of social integration
Berlin Aging Study (Smith & Baltes, 1997)
1 510 Cognition, self and personality, social integration
Health, mortality, sociodemographic, well-being
Cluster analysis Nine groups characterized by differing levels across all group defining domains
Epidemiological survey in Canberra (Jorm et al., 1998)
1 997 Cognition, health Health habits, personality, sociodemographic, verbal intelligence
Predefined by successful aging criteria
Three groups: successful aging, usual aging, and diseased aging
Berlin Aging Study (Smith & Baltes, 1998)
1 508 Cognition, health, self and personality, socioeconomic status, social integration, well-being
Age, gender Cluster analysis Eleven groups characterized by differing levels across all group defining domains; categorized by desirability level
(Appendix continues)
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2322 MORACK, RAM, FAUTH, AND GERSTORF
Appendix (continued)
Study Wave N Group defining domains External variables Grouping method Group findings
Data from Israeli Central Bureau of Statistics (Litwin, 2001)
1 2,079 Social network Morale, sociodemographic
Cluster analysis Five social network types
Australian Longitudinal Study of Aging (Andrews et al., 2002)
1 1,043 Cognition, functional health Health, lifestyle, psychological status, sociodemographic
Predefined by successful aging criteria
Three groups: higher, intermediate, and lower functioning
New England Centenarian Study (Evert et al., 2003)
1 424 Health Gender Predefined top-down procedure
Three groups characterized by age of onset of age-related illness
Americans’ Changing Lives Study (Fiori et al., 2006)
1 1,669 Social network Mental health, sociodemographic
Cluster analysis Five social network types
Berlin Aging Study (Fiori et al., 2007)
1 516 Social network Morbidity, sociodemographic, well-being
Cluster analysis Six social network types
Health and Aging Study (Ko et al., 2007)
1 287 couples
Cognition, health, personality, social support
Age, well-being Latent profile analysis
Two- and four-group solutions characterized by differing degrees of successful aging
Wisconsin Longitudinal Study (Fiori & Jager, 2012)
1 6,824 Social support Mental and physical health, sociodemographic
Latent class analysis Six groups: differing social support networks
Longitudinal studies Duke Longitudinal
Study of Aging (Manton et al., 1986)
11 267 Cognition Mental health, physical health, sociodemographic
Grade of membership model
Five groups characterized by differing cognitive abilities over time
Health-70 Study (Maxson et al., 1996)
3 335 Cognition, health, social contacts, well-being
Mortality, socioeconomic status, gender
Cluster analysis Five groups characterized by differing levels across all group defining domains
Wisconsin Longitudinal Study (Singer et al., 1998)
3 1,172 Depression, well-being Multiple life history variables
Predefined by history of depression and current well-being
Four groups: depressed/ unwell, healthy, vulnerable, and resilient
National Long-Term Health Survey (Manton & Land, 2000)
4 20.000 Functional health Medical conditions, mortality, gender
Grade of membership model
Seven groups characterized by disability state
Normative Aging Study (Aldwin et al., 2001)
4 1,515 Mental and physical health Health behaviors, mortality, personality, sociodemographic
Cluster analysis Four mental health groups and six physical health groups characterized by health trajectories
Betula Study (Lövdén et al., 2005)
3 500 Cognition Age, education, gender
Cluster analysis Six baseline groups characterized by differing cognitive levels; stable membership
Berlin Aging Study (Gerstorf et al., 2006)
3 132 Cognition, self and personality, social integration
Health, mortality, sociodemographic, well-being
Cluster analysis Three baseline groups characterized by differing levels across all group defining domains; fairly stable over time
National Longitudinal Survey of Mature Women (Wong & Hardy, 2009)
4 1,064 Retirement expectations Employment, health, sociodemographic
Latent class analysis Four groups characterized by retirement expectation patterns
(Appendix continues)
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2323PSYCHOLOGICAL PROFILES AND SUCCESSFUL AGING
Received September 5, 2012 Revision received December 4, 2012
Accepted December 12, 2012 �
New Editors Appointed, 2015–2020
The Publications and Communications Board of the American Psychological Association an- nounces the appointment of 6 new editors for 6-year terms beginning in 2015. As of January 1, 2014, manuscripts should be directed as follows:
● Behavioral Neuroscience (http://www.apa.org/pubs/journals/bne/), Rebecca Burwell, PhD, Brown University
● Journal of Applied Psychology (http://www.apa.org/pubs/journals/apl/), Gilad Chen, PhD, University of Maryland
● Journal of Educational Psychology (http://www.apa.org/pubs/journals/edu/), Steve Graham, EdD, Arizona State University
● JPSP: Interpersonal Relations and Group Processes (http://www.apa.org/pubs/journals/psp/), Kerry Kawakami, PhD, York University, Toronto, Ontario, Canada
● Psychological Bulletin (http://www.apa.org/pubs/journals/bul/), Dolores Albarracín, PhD, University of Pennsylvania
● Psychology of Addictive Behaviors (http://www.apa.org/pubs/journals/adb/), Nancy M. Petry, PhD, University of Connecticut School of Medicine
Electronic manuscript submission: As of January 1, 2014, manuscripts should be submitted electronically to the new editors via the journal’s Manuscript Submission Portal (see the website listed above with each journal title).
Current editors Mark Blumberg, PhD, Steve Kozlowski, PhD, Arthur Graesser, PhD, Jeffry Simpson, PhD, Stephen Hinshaw, PhD, and Stephen Maisto, PhD, will receive and consider new manuscripts through December 31, 2013.
Appendix (continued)
Study Wave N Group defining domains External variables Grouping method Group findings
Taiwan Longitudinal Survey on Aging (Hsu & Jones, 2012)
4–5 4,817 Chronic disease, depressive symptoms, economic satisfaction, physical functioning, social support and participation
Life satisfaction, self-rated health, sociodemographic
Multiple trajectory model analysis
Four successful aging groups each for a younger and an older cohort
Health and Retirement Study (Wickrama et al., 2012)
5 1,945 Depressive symptoms, memory problems, physical illness, physical impairment
Socioeconomic status (childhood and adult), sociodemographic
Latent trajectory class analysis
Three groups with different multidimensional health patterns
T hi
s do
cu m
en t
is co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
ti on
or on
e of
it s
al li
ed pu
bl is
he rs
. T
hi s
ar ti
cl e
is in
te nd
ed so
le ly
fo r
th e
pe rs
on al
us e
of th
e in
di vi
du al
us er
an d
is no
t to
be di
ss em
in at
ed br
oa dl
y.
2324 MORACK, RAM, FAUTH, AND GERSTORF
