Individual Differences in Rates of Change in Cognitive ... · Individual Differences in Rates of...

Individual Differences in Rates of Change in CognitiveAbilities of Older Persons

Robert S. Wilson, Laurel A. Beckett, Lisa L. Barnes, Julie A. Schneider,Julie Bach, Denis A. Evans, and David A. Bennett

Rush–Presbyterian–St. Luke’s Medical Center

The authors examined change in cognitive abilities in older Catholic clergy members. For up to 6 years,participants underwent annual clinical evaluations, which included a battery of tests from which summarymeasures of 7 abilities were derived. On average, decline occurred in each ability and was more rapid inolder persons than in younger persons. However, wide individual differences were evident at all ages.Rate of change in a given domain was not strongly related to baseline level of function in that domainbut was moderately associated with rates of change in other cognitive domains. The results suggest thatchange in cognitive function in old age primarily reflects person-specific factors rather than an inevitabledevelopmental process.

Longitudinal studies have shown that, on average, cognitiondeclines in old age (Colsher & Wallace, 1991; Evans et al., 1993;Wilson, Beckett, Bennett, Albert, & Evans, 1999). On an individ-

ual level, however, wide variability is seen (Christensen et al.,1999; Wilson et al., 1999). In some people cognition declinesprecipitously, but in many others cognition declines only slightlyor not at all, or improves slightly. Determining the factors thatcontribute to this variability is likely to require detailed knowledgeabout individual differences in patterns of change in differentcognitive abilities in old age. Unfortunately, such knowledge iscurrently limited, owing to several factors.

One factor is that few longitudinal studies have been designed tomodel change in individuals. The ability to do so depends not onlyon observing people for a sufficiently long period but also on thenumber and spacing of observations during that period. Yet mostlongitudinal studies of cognition in older people have only twoobservations, and few have more than three (e.g., Fozard, Ver-cruyssen, Reynolds, Hancock, & Quilter, 1994; Jacqmin-Gadda,Fabrigoule, Commenges, & Dartigues, 1997; Schaie, 1996;Siegler, 1983; Wilson et al., 1999). Such designs can yield accu-rate estimates of change at the group level, but they make itdifficult to reliably separate random variability from person-specific patterns of change. Further, intervals between observa-tions have ranged from about 6 months to well over a decade.However, cognition can decline rapidly in older persons, andindividuals experiencing such precipitous decline may be lost tofollow-up in studies with intervals of more than 3 or 4 yearsbetween observations (e.g., Gold et al., 1995; Owens, 1966;Schaie, 1996; Zelinski & Burnight, 1997).

A second issue is the extent to which some forms of cognitionare more likely to change in old age than are other forms. Incross-sectional studies, age-associated differences are typicallylarge in episodic memory and processing abilities such as workingmemory and perceptual speed but are minimal in crystallizedverbal abilities such as vocabulary and reading recognition (Horn,1968; Salthouse, 1988). Unfortunately, some longitudinal studieshave used only global measures of cognition, obscuring a potentialsource of individual differences (e.g., Albert et al., 1995; Brayne,Gill, Paykel, Huppert, & O’Connor, 1995; Deeg, Hofman, &van Zonneveld, 1990; Farmer, Kittner, Rae, Bartko, & Reiger,1995; Feskens et al., 1994; Izaks, Gussekloo, Dermout, Heeren, &

Robert S. Wilson and Lisa L. Barnes, Rush Alzheimer’s Disease Centerand Rush Institute for Healthy Aging, Department of Neurological Sci-ences and Department of Psychology, Rush–Presbyterian–St. Luke’s Med-ical Center, Chicago; Laurel A. Beckett and Denis A. Evans, Rush Alz-heimer’s Disease Center and Rush Institute for Healthy Aging, Departmentof Internal Medicine, Rush–Presbyterian–St. Luke’s Medical Center; JulieA. Schneider, Julie Bach, and David A. Bennett, Rush Alzheimer’s DiseaseCenter and Rush Institute for Healthy Aging, Department of NeurologicalSciences, Rush–Presbyterian–St. Luke’s Medical Center.

This research was supported by National Institute on Aging Grants RO1AG15819, K08 AG00849, and P30 AG10161. We are indebted for thealtruism and support of the hundreds of nuns, priests, and brothers from thefollowing groups participating in the Religious Orders Study: Archdioc-esan priests of Chicago, Dubuque, IA, and Milwaukee, WI; BenedictineMonks, Lisle, IL, and Collegeville, MN; Benedictine Sisters of Erie, Erie,PA; Benedictine Sisters of the Sacred Heart, Lisle; Capuchins, Appleton,WI; Christian Brothers, Chicago and Memphis, TN; Diocesan priests ofGary, IN; Dominicans, River Forest, IL; Felician Sisters, Chicago; Fran-ciscan Handmaids of Mary, New York; Franciscans, Chicago; Holy SpiritMissionary Sisters, Techny, IL; Maryknolls, Los Altos, CA, and Mary-knoll, NY; Norbertines, DePere, WI; Oblate Sisters of Providence, Balti-more; Passionists, Chicago; Presentation Sisters, Dubuque; Servites, Chi-cago; Sinsinawa Dominican Sisters, Chicago, and Sinsinawa, WI; Sistersof Charity, Blessed Virgin Mary, Chicago and Dubuque; Sisters of theHoly Family, New Orleans, LA; Sisters of the Holy Family of Nazareth,DesPlaines, IL; Sisters of Mercy of the Americas, Chicago, Aurora, IL, andErie; Sisters of St. Benedict, St. Cloud and St. Joseph, MN; Sisters of St.Casimir, Chicago; Sisters of St. Francis of Mary Immaculate, Joliet, IL;Sisters of St. Joseph of LaGrange, LaGrange Park, IL; Society of DivineWord, Techny, IL; Trappists, Gethsemane, KY, and Peosta, IA; WheatonFranciscan Sisters, Wheaton, IL. We also thank Todd Beck and WengingFan for statistical programming and Diane Atwater and Carolyn DeVivofor preparing the manuscript.

Correspondence concerning this article should be addressed to Robert S.Wilson, Rush Alzheimer’s Disease Center, 1645 West Jackson Boulevard,Suite 675, Chicago, Illinois 60612. E-mail: [email protected]

Psychology and Aging Copyright 2002 by the American Psychological Association, Inc.2002, Vol. 17, No. 2, 179–193 0882-7974/02/$5.00 DOI: 10.1037//0882-7974.17.2.179

179

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

Ligthart, 1995; Jacqmin-Gadda et al., 1997; Lyketsos, Chen, &Anthony, 1999; Unger, van Belle, & Heyman, 1999; Wilson et al.,1999).

Another issue is that because older people differ widely not onlyin level of cognitive function but also in rate of change, longitu-dinal studies require outcome measures that assess an especiallywide range of ability. Use of composite measures constructed fromindividual cognitive tests that vary in difficulty level can reducethe likelihood that floor or ceiling effects and other sources ofmeasurement error will distort estimates of change. Unfortunately,relatively few longitudinal studies have used composite cognitiveoutcome measures (but see Albert et al., 1995; Hultsch, Hertzog,Dixon, & Small, 1998; Malec, Smith, Ivnik, Peterson, & Tangalos,1996; Schaie, 1996; Wilson et al., 1999).

The ability to characterize individual differences in change incognitive function also depends on maintaining a high rate offollow-up participation among survivors. Because attrition inlongitudinal cognitive function studies is nonrandom and isgenerally found to be related not only to cognitive performancebut also to key predictors (e.g., age and education; Hultsch etal., 1998; Schaie, 1996; Siegler & Botwinick, 1979), it limitsboth the range of individual differences and the ability toidentify contributing factors. Yet rates of follow-up participa-tion vary widely in longitudinal studies or cannot be determinedfrom published information. The proportion of participantsavailable for longitudinal analyses is sometimes further reducedby analytic approaches that require equal numbers of observa-tions per individual (e.g., Colsher & Wallace, 1991; Emery,Pedersen, Svartengren, & McClern, 1998; Hopp, Dixon, Grut,& Backman, 1997; McCarty, Siegler, & Logue, 1982; B. J.Small, Dixon, Hultsch, & Hertzog, 1999).

In this study, we examined change in cognitive function over a6-year period in a large cohort of older persons. Participants wereCatholic clergy members who were 65 years and older and free ofclinical evidence of Alzheimer’s disease at baseline. They under-went annual clinical evaluations with over 95% participation infollow-up among survivors. Each evaluation included administra-tion of 21 cognitive function tests and clinical classification ofAlzheimer’s disease and other common conditions of old age.Based in part on a factor analysis of the baseline data, summarymeasures of seven cognitive abilities were formed.

Longitudinal analyses addressed four basic issues about cogni-tive function in old age. The first aim was to characterize individ-ual paths of change in each cognitive domain and then to examinethe variability in these paths. That is, in this relatively homoge-neous group of healthy, high-functioning older persons, does in-dividual cognitive change follow a regular path, suggesting ahomogeneous aging process, or are individual paths more varied?We used random-effects regression models to assess change. Inthis approach, change is modeled at both the individual level andthe group level. Change in an individual is assumed to follow themean path of change in the group, except for random effects thatcause the initial level of function to be higher or lower and the rateof change to be faster or slower. We used these random effects toestimate individual growth curves, which were then plotted andexamined for variability.

A second aim was to see how much of the variability in rates ofchange in cognitive function could be explained by knowledge ofa person’s baseline level of function. Most research on cognitive

function in old age is cross-sectional. It is important to know,therefore, how cross-sectional information about age-related dif-ferences in cognitive performance is related to observed differ-ences in rate of change. Information about this association islimited, however, because initial level of function has usually beenviewed only as a source of bias in estimating change. An advan-tage of the growth-curve approach used in this study is that initiallevel of function and rate of change are treated as separate randomeffects, and the correlation between them is explicitly estimated bythe models.

A third aim was to see the extent to which change in differentcognitive abilities covaried over the study period. Cross-sectional studies have indirectly examined this issue by com-paring the amount of age-related variance on tests of differentabilities and by estimating how much of that variance is shared,but interpretation of these data has varied (Baltes & Linden-berger, 1997; Lindenberger & Baltes, 1994; Rabbitt, 1993;Salthouse, Fristoe, & Rhee, 1996; Verhaegen & Salthouse,1997). Longitudinal studies can directly examine this issue, butfew have done so, and their results have been inconsistent. Inthe Victoria Longitudinal Study, rates of change over 6 years onnine composite measures of specific cognitive functions weremoderately intercorrelated, and a general cognitive change fac-tor accounted for most of the covariation on specific cognitivemeasures (Hultsch et al., 1998). By contrast, in the BaltimoreLongitudinal Study of Aging, rates of change in older personsover periods ranging from approximately 5 to 25 years onindividual tests of semantic memory (Wechsler Adult Intelli-gence Scale Vocabulary subtest) and episodic memory (BentonVisual Retention Test) were not significantly correlated (Aren-berg, 1978; Giambra, Arenberg, Zonderman, Kawas, & Costa,1995). In another longitudinal study, rates of change over 4years on tests of perceptual speed (Wechsler Adult IntelligenceScale Digit Symbol subtest) and episodic memory (Buschkeselective reminding) were not significantly correlated in a co-hort of 30 older persons (Taylor, Miller, & Tinklenberg, 1992).

The final aim of this study was to assess practice effects. Inlongitudinal studies, repeated experience (or practice) with cogni-tive tests often results in improved performance, making it difficultto know the actual level of change that has occurred. Manylongitudinal studies have used alternative forms in an effort toreduce practice effects (Giambra et al., 1995; Hertzog, Dixon, &Hultsch, 1992; Hultsch et al., 1998; Prince, Lewis, Bird, Blizard,& Mann, 1996; Youngjohn & Crook, 1993). Unfortunately, alter-native forms do not eliminate practice effects (Dikmen, Heaton,Grant, & Temkin, 1999; Hultsch et al., 1998; Watson, Pasteur,Healy, & Hughes, 1994), suggesting that factors other than learn-ing test content must be contributing (e.g., improved test-takingskills, reduced anxiety). In longitudinal studies, alternative formsadd another source of variability to individual paths of change,thereby complicating efforts to characterize how people differ. Inthis study, the same cognitive tests were administered at annualintervals. We constructed a series of models to see how thisexperience affected our estimates of the average level of change ineach domain and of how change in a given domain was related toinitial level of function in that domain and rate of change in otherdomains.

180 WILSON ET AL.

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

Method

Participants

Participants were part of the Religious Order Study, an ongoing longi-tudinal clinical–pathological study of aging and Alzheimer’s disease inolder Catholic nuns, priests, and brothers. Participants were recruited frommore than 30 groups across the country (see the author note). Eligibilityrequired (a) age of 65 years or older, (b) absence of evidence of dementiaor Alzheimer’s disease, and (c) consent to annual clinical evaluations andto brain donation at the time of death. Recruitment began in January 1994and is ongoing. The study was approved by the Human InvestigationsCommittee of Rush–Presbyterian–St. Luke’s Medical Center.

Analyses were based on 694 persons who completed the baseline clinicalevaluation between January 1994 and June 2000 and were found not tomeet clinical criteria for dementia or Alzheimer’s disease (see the ClinicalEvaluation section). At baseline, mean age was 75.9 years (SD � 6.9),mean education was 18.1 years (SD � 3.3), and mean score on theMini-Mental State Examination (MMSE; Folstein, Folstein, & McHugh,1975) was 28.3 (SD � 1.7). There were 443 women and 251 men; 652participants were White, 39 were Black, and 3 belonged to other racialgroups.

Clinical Evaluation

At baseline, each person had a structured, uniform evaluation, which wasrepeated annually with examiners who were unaware of previously col-lected data. The evaluation included a medical history, a neurologicalexamination, an assessment of cognitive function (see the Assessment ofCognitive Function section), and a review of a brain scan when available.Following this evaluation, a board-certified neurologist classified personswith respect to Alzheimer’s disease and other common neurologic disor-ders as previously described (Bennett, Shannon, Beckett, Goetz, & Wilson,1997; Kordower et al., 2001; Mufson et al., 1999). In short, the diagnosesof dementia and Alzheimer’s disease followed the criteria set forth by thejoint working group of the National Institute of Neurological and Com-municative Disorders and Stroke and the Alzheimer’s Disease and RelatedDisorders Association (NINCDS/ADRDA criteria; McKhann et al., 1984).At baseline, 48 people met NINCDS/ADRDA criteria for dementia (historyof cognitive decline and impairment in at least two cognitive domains), andthey are excluded from all analyses in this article. Of these 48 people, 39met NINCDS/ADRDA criteria for probable Alzheimer’s disease, and 9met criteria for possible Alzheimer’s disease because another conditionwas felt to contribute to cognitive impairment (3 with stroke and 2 withParkinson’s disease) or because memory was not impaired (n � 4).

Assessment of Cognitive Function

A battery of 21 cognitive function tests was administered at eachevaluation by examiners who had an average of 5 weeks of training andwere certified on each test using performance-based criteria. Data werecollected on laptop computers with forms programmed in Blaise (CentralBureau of Statistics, Voorburg, the Netherlands), a Pascal-based data-entryprogram, and scored in SAS (SAS Institute, 2000).

One test, the MMSE, provided a measure of global cognitive functionthat was used only for descriptive purposes. Scores on a test of auditorycomprehension, Complex Ideational Material (Goodglass & Kaplan,1983), were highly skewed, and this test was dropped from all subsequentanalyses, which were based on the remaining 19 tests.

Of 7 episodic memory tests, 4 involved story retention: immediate anddelayed recall of the East Boston Story (Albert et al., 1991) and of StoryA from Logical Memory of the Wechsler Memory Scale—Revised (Wechs-ler, 1987). Three tests involved learning and retaining a 10-word list(Morris et al., 1989): Word List Memory is the total number of wordsimmediately recalled after each of three consecutive presentations of the

list; Word List Recall is the number of words recalled after an approxi-mately 3-min delay; and Word List Recognition is the number of wordscorrectly recognized in a four-alternative, forced-choice format, adminis-tered after Word List Recall.

Four tests assessed semantic memory. Two measures required generat-ing words in response to pictorial or semantic stimuli. On a 20-item versionof the Boston Naming Test (Kaplan, Goodglass, & Weintraub, 1983),participants were asked to produce the names of pictured objects. VerbalFluency involved generating exemplars from two semantic categories(animals; fruits and vegetables) in separate 1-min trials. The total numberof unique exemplars was the performance measure. Two tasks assessedword knowledge. On the Extended Range Vocabulary Test (Ekstrom,French, Harman, & Kermen, 1976), participants were asked to select thebest definition for each target word from five alternatives; 15 itemsselected from Parts 1 and 2 of the original test were administered. On theReading Test, participants were asked to read aloud a series of 20 wordswith atypical spelling–sound correspondences (e.g., debt). The words wereselected from the National Adult Reading Test (Nelson, 1982) and itssubsequent modifications (Blair & Spreen, 1989; Grober & Sliwinski,1991).

Four measures of working memory were given. The Digit Span Forwardand Digit Span Backward subtests were administered and scored followingWechsler Memory Scale—Revised procedures (Wechsler, 1987). On DigitOrdering, participants were read sequences of two to nine digits, with twoitems at each length, and asked to say them back in ascending order. Itemswere scored as correct or incorrect, and administration was discontinuedwhen participants failed both items at a given length. The span format andbinary scoring of items are departures from the original version of this test(Cooper & Sagar, 1993; Cooper et al., 1992; Cooper, Sagar, Jordan,Harvey, & Sullivan, 1991). On Alpha Span, sequences of two to eightwords were presented, with two items at each length, and participants triedto say them back in alphabetical order (Craik, 1986). Administration wasdiscontinued when participants failed both items at a given length.

Two measures of perceptual speed were used. On the oral version of theSymbol Digit Modalities Test (Smith, 1982), participants had 90 s toidentify as many digit–symbol matches as possible. On Number Compar-ison (Ekstrom et al., 1976), participants classified pairs of 3- to 10-digitsequences as same or different. The performance measure was the numberof pairs correctly classified in 90 s minus the number incorrectly classified.

Two measures of visuospatial ability were administered. Judgment ofLine Orientation (Benton, Sivan, Hamsher, Varney, & Spreen, 1994)required participants to estimate the angle subtended by two lines in amatch-to-sample format. Fifteen of the original 30 items were used. On theStandard Progressive Matrices (Raven, Court, & Raven, 1992), participantswere shown an abstract visual display with one element missing and askedto select the missing element from an array of six to eight alternatives.Seventeen items were used from a total of 60 in the original test.

To enhance the applicability of the visual tests with older participants,stimuli were enlarged for the Boston Naming Test, Symbol Digit Modal-ities Test, Number Comparison, Judgment of Line Orientation, and Stan-dard Progressive Matrices.

Data Analysis

We first obtained univariate descriptions of the scores, both graphicaland computed, for each test in each year. Because some of these distribu-tions were relatively skewed and because of the importance of minimizingfloor and ceiling effects in longitudinal analyses, we used composites of 2or more tests as outcomes. We developed these composite measures in athree-step process. First, we specified two conceptually based groupings ofthe tests into functional domains. Second, we developed an empiricallybased grouping of the tests by (a) performing a principal-componentsfactor analysis, with varimax rotation, of the 19 tests at baseline and (b)grouping tests together with rotated factor loadings of .50 or higher on a

181CHANGE IN COGNITIVE ABILITIES OF OLDER PERSONS

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

common factor. Any test without a factor loading of .50 or higher wasgrouped empirically into its own one-test domain. We used baseline data todevelop the empirical grouping because the cohort was most disease free atthis point. Third, to assess the agreement of each conceptually basedgrouping with the empirically based grouping, we used Rand’s statistic(Rand, 1971), rescaled from �1 to �1 to be comparable to Kendall’s tau,another measure that calculates proportion of pairs concordant minusproportion discordant. Two groupings were considered to agree on a pairof tests if both groupings put the tests in the same cluster or if bothgroupings put the tests in different clusters; two groupings were consideredto disagree if one grouping put a pair of tests in the same cluster and theother grouping put them in different clusters (Bennett, Shannon, Beckett, &Wilson, 1999). We scored the agreement of each conceptual model withthe data-driven grouping across all pairs of tests, scoring agreement as �1and disagreement as �1 and averaging the scores to yield an overall indexof agreement. This index can range from perfect agreement (�1) to perfectdisagreement (�1). Strong agreement between the data-driven groupingand a given conceptual model was taken as support for using the concep-tual domains in analyses. We used a permutation test to see if a given levelof agreement could have been obtained by chance alone. We calculated theagreement that would be obtained for all groupings of the same pattern asthe empirical grouping (number of groups, number of tests per group) forall possible permutations of the tests. We then computed what fraction ofthe permutations gave agreement with the conceptual grouping as high asthat from the empirical grouping, and this fraction formed the p value basedon the permutation test. An advantage of this approach is that it does notassume multivariate normality, which was an important considerationgiven the negatively skewed distributions of some individual test scores atbaseline.

The next goal of the analysis was to characterize change in cognitivefunction over the study period within each of the domains. We used agrowth-curve approach, fitting random-effects regression models (Laird &Ware, 1982). These models enabled us to estimate the mean path of changein each domain and to characterize how individual paths differed from themean trajectory. This approach has been used in several previous longitu-dinal studies of cognitive function (Berg et al., 1992; Hebert et al., 2000;Jacqmin-Gadda et al., 1997; Teri, Hughes, & Larson, 1990; Wilson et al.,1999; Wilson, Bennett, Gilley, Beckett, Barnes, & Evans, 2000; Wilson,Bennett, Gilley, Beckett, Schneider, & Evans, 2000b; Wilson, Gilley,Bennett, Beckett, & Evans, 2000a, 2000b) and other behavioral outcomes(Gibbons et al., 1993; Wilson, Bennett, Gilley, Beckett, Schneider, &Evans, 2000a). Practical advantages of this approach are that individualsneed not have the same number of observations and the time betweenobservations is not assumed to be constant within or between persons.

The core model had three predictors: age (at baseline), time (in years)since baseline, and their interaction. To ease interpretation of results, wecentered age at 75 years, the round number closest to the mean age of thecohort, which means that the reference person for analyses is a 75-year-old.The coefficient for age indicates the average effect on cognitive score ofbeing 1 year older (or younger) than 75 at baseline, which can be thoughtof as the cross-sectional effect of age. The coefficient for time indicates theaverage effect of 1 year of study time on cognitive score for a 75-year-old,which can be viewed as the longitudinal effect of age. The interaction ofage and time indicates the extent to which annual rate of change varies witheach year of baseline age above or below 75. In secondary analyses, weadded a quadratic term for age to the core model to see if the associationof age with cognitive function at baseline had a nonlinear component.

To assess heterogeneity in individual paths of change, we examined thedistribution of the model estimates of the individual slopes in each cogni-tive domain. We also used the estimated intercepts and slopes to constructindividual growth curves, which were plotted.

The model also provided an estimate of the correlation between theperson-specific intercepts and slopes in each domain, which we used to seehow much of the variability in annual change between persons could be

accounted for by differences in baseline level of function. To assess thecovariation in change among different cognitive abilities, we computed thecorrelations among the person-specific slopes in each domain of functionand performed a principal-components factor analysis on these slopes.

The final analytic goal was to examine practice effects, which wedefined as changes in performance that could be attributed to prior expe-rience with the tests and the testing situation. Because previous researchsuggests that most of the benefits of practice result from initial testingexperiences (Jacqmin-Gadda et al., 1997; Unger et al., 1999), we initiallyconstructed a random-effects model, which we designated as Model 1, withterms for time; for the contrast between the baseline and the mean of allfollow-up evaluations, to estimate the average effect of the first practiceexperience (at baseline); and for the contrast between the first follow-upand the mean of all subsequent follow-up evaluations, to estimate theadditional average effect of the second practice experience.

We further examined the effects of practice in secondary analyses. First,we added a second follow-up term to Model 1 to assess the incrementaleffect of the third testing experience. Second, we constructed another set ofrandom-effects models, each with a term for time. We then compared thecoefficients for time from these models with those obtained from Model 1to see how practice influenced our estimate of the average annual changein each domain. Third, we used person-specific intercepts and slopes fromModel 1 to estimate the correlations of initial level with rate of change ineach domain and the correlations among rates of change in differentdomains. We then compared these correlations with those obtained fromthe core model to see how practice influenced our estimates of theseassociations.

We used SAS PROC MIXED (SAS Institute, 2000) to estimate theparameters of the random-effects models. The models assume a linearassociation of predictors with outcomes. We checked this assumptionanalytically, by adding a quadratic term for age to the core model, andgraphically, by plotting residuals against age. We assessed model assump-tions about normality, independence of errors, and homoscedasticity oferrors by plotting the estimated intercepts by slopes, by plotting residualsfrom each time point, and by examining selected summary statistics. Weconcluded that model assumptions were adequately met.

Results

Development of Cognitive Function Measures

Table 1 shows the baseline mean and standard deviation of eachof the 19 cognitive function tests used in analyses. Four tests(Immediate and Delayed Story Recall, Word List Recognition, andBoston Naming Test) had negatively skewed distributions, withmany scores at or near the ceiling. The distributions of the other 15test scores were relatively symmetric.

To construct composite measures from these individual tests foruse in longitudinal analyses, we began by hypothesizing two waysof grouping tests into functional domains (Groupings 1 and 2 inTable 1). In one grouping, we specified five domains: episodicmemory, semantic memory, working memory, perceptual speed,and visuospatial ability. In the second grouping, we specifiedseven domains. We subdivided episodic memory tests into story-retention and word-retention clusters because some longitudinalstudies have reported that practice effects are more pronounced onthe former than the latter (McCarty et al., 1982; B. J. Small, Dixon,et al., 1999). We subdivided semantic memory tests into word-generation and word-knowledge clusters because of cross-sectional evidence that age-related differences are more pro-nounced on the former than the latter (Burke & Peters, 1986;MacKay, Abrams, & Pedroza, 1999; Nelson, 1982; Salthouse,1993; Troyer, Moscovitch, & Winocur, 1997).

182 WILSON ET AL.

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

We next developed an empirically based grouping of the tests byperforming a principal-components factor analysis (with varimaxrotation) on the 19 tests at baseline. The analysis, summarized onthe right side of Table 1, identified five factors. We then groupedtogether the tests that loaded .50 or higher on a common factor. Asshown in Table 1, the grouping based on the factor analysis wassimilar to both of the hypothesized groupings. We calculated anindex of agreement between the two groupings using Rand’smeasure, rescaled to range from �1 (complete discordance be-tween the two groupings) to �1 (complete concordance betweenthe two groupings). The overall agreement of the factor analyticgrouping with the five-domain grouping was .79 and with theseven-domain grouping was .84 (both ps � .01), providing em-pirical support for both hypothesized groupings. We chose theseven-domain grouping because the fit was slightly better andbecause we wanted to examine change within the subdivisions ofepisodic and semantic memory.

Summary measures of each of the seven cognitive domains wereformed, as in previous research (Wilson et al., 1999, 2000), byconverting raw scores on each component test to z scores, using thebaseline mean and standard deviation, and then averaging the zscores of the component tests in that domain. If component scoreswere missing, the domain score was missing unless at least half ofthe component tests had valid scores, in which case the domainscore was based on the valid test scores. Table 2 shows thebaseline mean, standard deviation, and range for the seven cogni-tive domain scores.

Follow-Up Participation

Of the 694 eligible participants, 25 died before the firstfollow-up evaluation, and another 57 had not yet reached their firstfollow-up date. Therefore, 612 persons were eligible for follow-up

at the time of these analyses. Information about follow-up partic-ipation is summarized in Table 3. Longitudinal analyses requireda minimum of two valid scores per individual, and 596 to 601persons met this criterion for each of the cognitive domain scores,over 97% of those eligible for follow-up. Among those individualswith follow-up data, there was an average of more than fiveobservations per individual (range two to seven), which repre-sented over 96% of possible scores.

Average Paths of Change

Random-effects models were used to summarize cross-sectionaland longitudinal information about the average paths of change ineach cognitive domain. The primary model included terms for age(at baseline), study time (in years), and their interaction. The termfor age is a cross-sectional estimate of the effect of being a yearolder at baseline. The term for time is a longitudinal estimate of the

Table 1Psychometric Information on the 19 Cognitive Function Tests at Baseline

Test M SD

Hypothesized cognitive domaina Factor loadingb

Grouping 1 Grouping 2 1 2 3 4 5

Logical Memory Ia 11.8 3.9 Episodic memory Story retention .52 .06 .06 .39 .46Logical Memory IIa 10.0 4.2 Episodic memory Story retention .62 .11 .09 .34 .42Immediate story recall 9.8 1.7 Episodic memory Story retention .11 .14 .11 .07 .87Delayed story recall 9.4 1.9 Episodic memory Story retention .21 .13 .15 .00 .84Word List Memory 17.7 4.1 Episodic memory Word retention .76 .26 .24 .04 .03Word List Recall 5.6 2.1 Episodic memory Word retention .82 .13 .12 .06 .10Word List Recognition 9.6 1.0 Episodic memory Word retention .69 .00 �.01 �.08 .11Boston Naming Test 18.1 2.0 Semantic memory Word generation .23 .38 .10 .47 .09Verbal Fluency 35.0 9.1 Semantic memory Word generation .45 .41 .27 .09 .11Extended Range Vocabulary Test 10.9 3.3 Semantic memory Word knowledge .04 .13 .24 .82 .07Reading test 13.6 4.1 Semantic memory Word knowledge .00 .09 .26 .81 .04Digit Span Forward 8.2 1.9 Working memory Working memory .02 .05 .78 .18 .10Digit Span Backward 6.3 2.0 Working memory Working memory .06 .11 .75 .19 .12Digit ordering 7.7 1.7 Working memory Working memory .15 .41 .50 .15 .10Alpha span 5.0 1.7 Working memory Working memory .27 .23 .70 .13 .04Symbol Digit Modalities Test 39.5 10.8 Perceptual speed Perceptual speed .25 .79 .19 �.01 .12Number Comparison 25.2 7.1 Perceptual speed Perceptual speed .07 .77 .13 .00 .10Judgment of Line Orientation 9.9 3.2 Visuospatial ability Visuospatial ability �.09 .48 .03 .30 .08Standard Progressive Matrices 10.3 3.4 Visuospatial ability Visuospatial ability .22 .62 .13 .31 .03

a We hypothesized that individual tests could be grouped into five (Grouping 1) or seven (Grouping 2) functional domains. b Factor loadings are fromprincipal-components analysis with varimax rotation; loadings of .50 or higher are in boldface.

Table 2Psychometric Information on the Cognitive Domain Measuresat Baseline

Cognitive domain M SD Range

Story retention 0.150 1.486 �4.256 to 3.733Word retention 0.153 1.228 �8.017 to 2.716Word generation 0.161 1.026 �4.075 to 4.427Word knowledge 0.065 1.263 �4.770 to 1.470Working memory 0.128 1.423 �4.420 to 4.261Perceptual speed 0.079 1.216 �4.708 to 3.737Visuospatial ability 0.086 1.139 �4.144 to 2.343

Note. Each cognitive domain measure is the mean of the z scores onindividual tests grouped in the domain.


This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

effect of getting a year older during the study. Their interactiondenotes whether cognitive change per year varies with baselineage.

Separate analyses were conducted for each of the cognitivedomain measures (see Table 4). On average, performance in eachdomain declined over the study period (as shown by the terms fortime). Thus, the story-retention measure declined an averageof 0.030 units per year (95% confidence interval [CI]: �0.045 to�0.016). There was also an inverse relation between age andbaseline score on each measure. Thus, the baseline story-retentionscore was reduced by an average of 0.031 units for each additionalyear of baseline age beyond 75 (95% CI: �0.039 to �0.023). Theaverage change in baseline score per year of age was roughlycomparable to the average annual decline observed over the studyperiod. That is, in each domain except working memory, the 95%CIs for the effects of a year of age at baseline and a year ofobservation in the study overlapped, although the degree of over-lap was minimal for word generation and word knowledge.

The average rate of change in each cognitive domain was relatedto baseline age (as shown by the interaction terms in Table 4). Onthe story-retention measure, for example, the average annual rateof decline increased by 0.008 units for each additional year ofbaseline age beyond 75. In a typical 85-year-old participant, there-fore, story retention declined 0.110 units annually compared withan average loss of 0.030 units per year in a 75-year-old and anaverage gain of 0.050 units per year in a 65-year-old.

To see if the association of age with cognitive function atbaseline also had a nonlinear component, we repeated the random-effects models, adding a quadratic term for age. This term wassignificant for word knowledge (parameter estimate � �0.002;95% CI: �0.003 to �0.001) but not for the other cognitive domainmeasures, indicating that the association of age and cognition atbaseline was mostly linear, unlike the association of age with rateof change in cognitive function.

Figures 1 and 2 summarize the cross-sectional and longitudinaleffects of age in each cognitive domain. Figure 1 shows baselinescores in each domain plotted by age at baseline, with a linearregression line added. In each domain, there was an approximatelylinear association between lower performance and increased age, withcorrelations ranging from �.10 (word knowledge) to �.45 (wordgeneration). A slight ceiling effect on the word-knowledge measurecan be seen in Figure 1, and this effect probably limited its linearassociation with age. For Figure 2, we divided the age range into5-year intervals and then estimated the average change during the

study for a typical participant followed for 5-years from the beginningof a given interval. In contrast to the mistaken interference about rateof change that might be drawn from the cross-sectional results shownin Figure 1 (that change is similar across age), Figure 2 demonstratesthat the average rate of decline in each domain increased with in-creasing baseline age. In each domain, little decline was observed in65- and 70-year-old persons, whereas substantial decline was seen in80- and 85-year-old persons.

Heterogeneity in Paths of Change

To examine individual patterns of change, we constructed mod-els with terms for time for each cognitive domain score. We firstestimated each person’s slope (i.e., annual rate of change) in eachdomain. The distributions of these slopes, shown in Figure 3, wereroughly symmetric and suggest considerable variability amongpersons. The smoothed person-specific curves estimated fromthese models are shown in Figure 4. So that individual paths couldbe distinguished, the plots were constructed for a 25% randomsample of the cohort. Figure 4 is organized according to the age ofthe participant at each evaluation; the length of each line relativeto the x-axis indicates the years of observation on that individual.Overall, substantial heterogeneity is evident in each cognitivedomain, with a few people exhibiting a precipitous decline andmost people remaining stable or declining or improving slightly.

Table 4Summary of Random-Effects Models Evaluating the Effects ofAge, Time, and Their Interaction on the CognitiveDomain Measures

Cognitive domainand model term Parameter estimate 95% confidence intervala

Story retentionAge �0.031 �0.039 to �0.023Time �0.030 �0.045 to �0.016Age � Time �0.008 �0.010 to �0.005

Word retentionAge �0.039 �0.046 to �0.032Time �0.042 �0.059 to �0.026Age � Time �0.010 �0.012 to �0.007

Word generationAge �0.044 �0.051 to �0.037Time �0.066 �0.082 to �0.051Age � Time �0.007 �0.010 to �0.005

Word knowledgeAge �0.015 �0.024 to �0.006Time �0.026 �0.037 to �0.015Age � Time �0.006 �0.008 to �0.004

Working memoryAge �0.030 �0.038 to �0.023Time �0.056 �0.067 to �0.046Age � Time �0.005 �0.006 to �0.003

Perceptual speedAge �0.056 �0.065 to �0.047Time �0.073 �0.085 to �0.061Age � Time �0.007 �0.009 to �0.005

Visuospatial abilityAge �0.046 �0.054 to �0.038Time �0.027 �0.038 to �0.015Age � Time �0.003 �0.005 to �0.002

a The 95% confidence interval is the range within which the true value ofthe parameter is 95% likely to be.

Table 3Information on Follow-Up Participation

Cognitive domain

No. with2 or more

scores

% with2 or more

scores

Meanno. ofscores

% ofpossiblescores

Story retention 600 98.0 5.3 98.0Word retention 600 98.0 5.3 97.8Word generation 599 97.9 5.3 98.2Word knowledge 601 98.2 5.3 98.3Working memory 600 98.0 5.3 98.2Perceptual speed 598 97.7 5.3 97.4Visuospatial ability 596 97.3 5.2 96.9

184 WILSON ET AL.

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

Although decline in each domain appears to be more common inolder persons than in younger persons, variability is evident at allages. Figure 4 also illustrates a relative absence of floor and ceilingeffects, except for a ceiling effect on the word-knowledge measure(see also Figure 1).

It also appears in Figure 4 that baseline level of function is notstrongly related to rate of change. To further investigate this effect, wecomputed the correlations of the person-specific intercepts (i.e., initiallevels of function) with the person-specific slopes in each domain.These correlations were as follows: word retention (r � .53), story

retention (r � .33), perceptual speed (r � .31), word generation (r �.29, p � .01), working memory (r � .14, p � .05), word knowledge(r � .04, p � .59), and visuospatial ability (r � .02).

Covariation in Rates of Change Across Domains

To see how change in one domain was related to change in otherdomains, we constructed a random-effects model (with a term fortime) for each cognitive domain measure, estimated the person-specific slopes in each domain, and then computed the correlations

Figure 1. Score in each cognitive domain at baseline plotted by age at baseline, with a linear regression lineadded.


This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

among the slopes (see Table 5). The correlations ranged from .37to .78, with a median of .54 (all ps � .01). We then submitted theslopes from each domain to a principal-components factor analy-sis. Each measure loaded on a single factor that accounted for61.6% of the variance among the slopes (see Table 5).

To determine whether the correlation between rates of changedepended on age, we constructed another set of random-effectsmodels with terms for age, time, and their interaction; estimatedthe slopes in each cognitive domain; and then repeated the factoranalysis. Results were similar to the initial analysis, with each

cognitive domain measure loading on a single factor that ac-counted for 56.3% of the variance, suggesting that age per se wasnot responsible for the observed covariation.

Practice Effects

Figures 3 and 4 show that the cognitive test performance of someparticipants improved over the study period. To determine the extentto which improvement reflected a practice effect, we constructedanother set of random-effects models (see Table 6, Model 1). Each

Figure 2. Predicted 5-year paths of change in each cognitive domain for typical participants who were 65,70, 75, 80, or 85 years old.

186 WILSON ET AL.

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

model had terms for time, for the contrast between the baseline and allsubsequent evaluations (to assess the average effect of having expe-rienced the testing at baseline), and for the contrast between the firstfollow-up and all subsequent follow-up evaluations (to assess theadditional average effect of the first follow-up testing experience).The baseline term was significant (and negative) for each cognitivedomain measure, consistent with the idea that the baseline testingexperience slightly boosted performance on follow-up evaluations.The first follow-up term was significant (and negative) for three of thecognitive measures (word generation, perceptual speed, and visuospa-

tial ability), suggesting that the second testing experience also boostedsubsequent performance, though effects were less than those associ-ated with the baseline testing experience. We added a secondfollow-up term in another set of analyses (data not shown). It wassignificant (and negative) for only two of the cognitive measures(word generation and perceptual speed), suggesting that the benefit ofpractice gradually diminished and varied somewhat among cognitivemeasures.

To see how practice affected our estimates of the mean annualchange in each measure, we compared the effects of time in the

Figure 3. Frequency histograms of the person-specific slopes in each cognitive domain estimated from arandom-effects model with a term for time.


This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

aforementioned analyses with the effects of time in another set ofmodels that did not include terms for baseline or follow-up (seeTable 6, Model 2). In each cognitive domain, the average decline peryear (as shown by the estimate for time) was greater in Model 1 thanin Model 2, indicating that our estimates of the average rate of declineover the full study period were reduced somewhat by the tendency ofparticipants to improve over the initial evaluations. That is, account-ing for practice in the model increased our estimate of the averageannual rate of decline in each domain.

To see whether practice contributed to the association betweenrates of change in different abilities, we submitted the person-specific slopes estimated from Model 1 to a principal-componentsfactor analysis. Results were similar to previous analyses, witheach cognitive measure loading on a single factor that accountedfor 61.8% of the variance. To see whether practice affected theassociation of initial level of function with rate of change, wecomputed the correlations between the intercepts and the slopesfrom Model 1. The correlations, which ranged from .00 to .51

Figure 4. Person-specific paths of change in each cognitive domain estimated for a 25% random sample of thecohort from a random-effects model with a term for time.

188 WILSON ET AL.

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

(median r � .27), were similar to those from random-effectsmodels that included a term only for time.

Discussion

In this study, we examined change in cognitive abilities over a6-year period in a large cohort of Catholic clergy who were 65

years and older and free of clinical evidence of Alzheimer’sdisease at baseline. There were three main findings. First, therewas substantial heterogeneity in individual rates of change over thestudy period. Some persons declined sharply in cognition, somedeclined gradually, and many others stayed the same or improved.Second, we found that baseline level of function in a givencognitive domain accounted for a relatively modest amount of the

Table 5Correlations Among Person-Specific Slopes in Each Cognitive Domain and Their Loadings on aCommon Factor

1 2 3 4 5 6 7Factor

loadinga

1. Story retention — .78 .70 .44 .60 .56 .48 .852. Word retention — .74 .44 .57 .61 .52 .873. Word generation — .46 .57 .65 .55 .874. Word knowledge — .41 .44 .37 .635. Working memory — .48 .43 .746. Perceptual speed — .54 .797. Visuospatial ability — .70

a Factor loadings are from a principal-components analysis of the person-specific slopes in each cognitivedomain.

Table 6Summary of Random-Effects Models Evaluating Practice Effects on the CognitiveDomain Measures

Cognitive domain andmodel terma

Model 1 Model 2

Estimate 95% CIb Estimate 95% CIb

Story retentionTime �0.068 �0.090 to �0.047 �0.038 �0.053 to �0.022Baseline �0.189 �0.260 to �0.117Follow-up �0.002 �0.061 to 0.056

Word retentionTime �0.092 �0.115 to �0.069 �0.051 �0.068 to �0.034Baseline �0.226 �0.295 to �0.156Follow-up �0.054 �0.111 to �0.003

Word generationTime �0.105 �0.126 to �0.084 �0.073 �0.089 to �0.057Baseline �0.161 �0.223 to �0.098Follow-up �0.077 �0.128 to �0.026

Word knowledgeTime �0.050 �0.067 to �0.033 �0.031 �0.043 to �0.020Baseline �0.103 �0.164 to �0.042Follow-up �0.031 �0.080 to 0.019

Working memoryTime �0.072 �0.088 to �0.056 �0.061 �0.072 to �0.050Baseline �0.068 �0.126 to �0.010Follow-up �0.007 �0.055 to 0.040

Perceptual speedTime �0.108 �0.127 to �0.090 �0.080 �0.093 to �0.067Baseline �0.144 �0.207 to �0.080Follow-up �0.073 �0.125 to �0.021

Visuospatial abilityTime �0.046 �0.064 to �0.028 �0.030 �0.041 to �0.018Baseline �0.076 �0.143 to �0.009Follow-up �0.058 �0.113 to �0.003

Note. Model 1 contained three terms, and Model 2 contained one term (time). CI � confidence interval.a The baseline term is for the contrast between the baseline and all subsequent evaluations; the follow-up termis for the contrast between the first follow-up and all subsequent evaluations. b The 95% CI is the range withinwhich the true value of the parameter is 95% likely to be.


This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

variability in rates of change in that domain. Third, rates of changein different cognitive domains were moderately correlated; that is,individuals who were declining in one domain were likely to bedeclining in other domains as well. Overall, the results suggest thatchange in cognitive function in old age is highly specific to theindividual.

Wide individual variation in paths of cognitive change has beenpreviously reported (Christensen et al., 1999; Hultsch et al., 1998;Wilson et al., 1999). It is especially noteworthy in the presentstudy because the detailed cognitive function histories available onparticipants made it possible to reliably characterize individualpatterns of change in diverse domains of cognitive function. Theobserved heterogeneity in rates of change is not easily reconciledwith the idea that cognitive ability declines inevitably or uniformlyin old age as part of some developmental process. The results aremore consistent with the accumulation of common age-relatedconditions that can affect multiple cognitive systems, for instance,Alzheimer’s disease (Haxby et al., 1990; Wilson et al., 2000) andcerebrovascular disease (Rockwood et al., 2000; Zhu et al., 1998).More definitive conclusions about the bases of age-related changein cognitive function will require better understanding of howdisease is related to person-specific change in cognition and howenvironmental and genetic factors modify this association. Be-cause all participants in this study have agreed to brain donation atthe time of death, we will have the opportunity to examine therelation of disease pathology to person-specific change in cogni-tive function.

Given that individuals in this cohort are changing substantiallybut at varying rates, an important issue becomes the extent towhich baseline level of function can predict change. Previouslongitudinal research provides little information about the associ-ation of initial level of cognition with rate of change (Alder, Adam,& Arenberg, 1990; Wilson et al., 1999), in part because moststudies have only two data points, making it difficult to disentanglestarting level from change. Another issue is that floor and ceilingeffects distort the association of initial level with change. As aresult, initial level of function has often been viewed only as asource of error in estimating change rather than as an additionalsource of information about how people change (Cronbach &Furby, 1970; Rogosa, 1988). In this cohort, we found that initiallevel of cognitive function was associated with rate of change, butnot strongly so. Thus, the person-specific intercepts accounted forless than 11% of the variance in the person-specific slopes in eachdomain except story retention, for which the shared variance wasapproximately 28%. In a previous study of a population-basedsample of older persons, Wilson et al. (1999) found a similarcorrelation (r � .36) between baseline score on a global cognitivemeasure and rate of change over 3.5 years of follow-up. Moreresearch on this complex issue is needed, but these results suggestthat cross-sectional data can provide a relatively limited amount ofinformation about individual differences in rate of cognitivechange in old age.

A long-standing question about cognitive function in old age iswhether age affects cognitive abilities in a global or selectivefashion, or to quote Rabbitt (1993), “Does it all go together whenit goes?” (p. 385). Most research on this issue has been cross-sectional, but our finding that individual differences at a givenpoint in time are not strongly related to individual differences inchange casts doubt on this approach. Unfortunately, as reviewed

above, longitudinal evidence about covariation in rates of changein different cognitive abilities is sparse and contradictory. In thisstudy, all forms of cognitive function declined during the 6 yearsof observation, which is generally consistent with previous longi-tudinal studies in this age range with at least 5 years of observationand multiple cognitive outcomes (e.g., Hultsch et al., 1998; Schaie,1996). Rates of decline in different cognitive domains were mod-erately correlated and loaded on a single factor in a principal-components analysis, and in secondary analyses, these findingswere not explained by age or practice effects. These results aresimilar to those obtained in the Victoria Longitudinal Study usingstructural equation modeling with composite cognitive measures(Hultsch et al., 1998). By contrast, two studies that used individualtests as outcomes did not find rates of change to be correlated(Giambra et al., 1995; Taylor et al., 1992). Further research isneeded, but we think it is likely that change in cognition in old ageis mostly global and that the negative results obtained by Giambraet al. and Taylor et al. reflect the psychometric limitations ofindividual cognitive tests as outcomes in longitudinal studies.

On average, performance on cognitive function tests benefitedfrom previous experience, or practice, with the tests, consistentwith prior studies with retest intervals of 1 year or longer (Dikmenet al., 1999; Hultsch et al., 1998; Jacqmin-Gadda et al., 1997;Mitrushina & Satz, 1991; Paolo, Troster, & Ryan, 1997; S. A.Small, Stern, Tang, & Mayeux, 1999; Unger et al., 1999). Most ofthe effect in this cohort was seen following the initial experiencewith the tests (at baseline). Lower levels of benefit were also seenon a subset of measures after the first and second follow-upevaluations. It is likely, therefore, that the observed levels ofdecline in cognitive test performance underestimate the actuallevels of decline in the abilities being measured.

Because the effects of practice, like those of age and time, areprobably heterogenous, practice also has the potential to affectindividual differences in rate of change. However, because theability to benefit from practice is probably related to level ofcognitive function, it seems unlikely that practice would substan-tially diminish individual differences in rates of change. We foundno evidence in this cohort that practice affected the covariation inrates of change in different domains or the association of initiallevel of function and rate of change within domains.

Rate of cognitive change was related to age, consistent withlongitudinal population-based studies of older persons (e.g.,Brayne et al., 1995; Colsher & Wallace, 1991; Evans et al., 1993;Izaks et al., 1995; Wilson et al., 1999). In each cognitive domain,older persons experienced more cognitive decline than youngerpersons. In story retention, for example, the score of an average65-year-old person improved 0.05 units per year, compared withan annual loss of 0.03 units in the average 75-year-old participantand an annual loss of 0.11 units in the average 85-year-old par-ticipant. This nonlinear increase in the average rate of decline withadvancing age contrasts with the linear association of age withcognitive function at baseline in this study. The latter result sug-gests that the range of cognitive function in this cohort may havebeen somewhat restricted at baseline, perhaps reflecting a healthyvolunteer effect or the exclusion of those individuals with demen-tia at baseline. It is likely, therefore, that the cross-sectional resultsalso underestimate the overall level of age-associated cognitivedecline in this cohort.

190 WILSON ET AL.

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

Several features of the present study are notable. First, on thebasis of a standard clinical evaluation and application of widelyaccepted criteria, persons with clinical evidence of Alzheimer’sdisease or other forms of dementia were identified at baseline andexcluded from analyses. Second, follow-up participation in survi-vors was very high, greatly reducing the chance that attrition couldbias estimates of change. Third, composites of two or more indi-vidual cognitive function tests were used as outcomes, reducingthe opportunity for floor and ceiling effects and other sources ofmeasurement error to distort estimates of change. Fourth, multipledomains of cognitive function were assessed, making it possible tocompare age effects across domains. Fifth, there was an average ofmore than five evenly spaced observations per person, which madeit possible to reliably characterize change in individuals.

These findings have important limitations. First, because thecohort was selected, results may not generalize to other groups. Onaverage, participants differed from older people in the U.S. pop-ulation in education (with an average of more than 18 years ofschooling) and lifestyle (all members of the Catholic clergy), andperhaps in other ways as well. Although this homogeneity mayhave helped to highlight age-associated effects, it is unlikely thatthe full spectrum of cognitive function was represented at baseline,and it is possible that heterogeneity in rates of change was alsounderestimated. Longitudinal cognitive function studies of morediverse cohorts are needed. Second, it is uncertain how to bestconstruct composite cognitive function measures for longitudinalanalyses. We based decisions on how to group individual tests onbaseline data alone because we wanted to assess covariation inchange in different functions and felt that our ability to do so mightbe compromised if follow-up data were used to guide groupingdecisions. We acknowledge, however, that longitudinal data can beused to determine groupings and that such an approach might yielddifferent results. Further research on this issue is needed. Third, alonger period of observation would improve our ability to charac-terize individual paths of change in cognitive function, especiallygradual or nonlinear patterns.

References

Albert, M. S., Jones, K., Savage, C. R., Berkman, L., Seeman, T., Blazer,D., & Rowe, J. W. (1995). Predictors of cognitive change in old persons:MacArthur studies of successful aging. Psychology and Aging, 10,578–589.

Albert, M., Smith, L., Scherr, P., Taylor, J., Evans, D., & Funkenstein, H.(1991). Use of brief cognitive tests to identify individuals in the com-munity with clinically diagnosed Alzheimer’s disease. InternationalJournal of Neuroscience, 57, 167–178.

Alder, A. G., Adam, J., & Arenberg, D. (1990). Individual-differencesassessment of the relationship between change in and initial level ofadult cognitive functioning. Psychology and Aging, 5, 560–568.

Arenberg, D. (1978). Differences and changes with age in the BentonVisual Retention Test. Journal of Gerontology, 33, 534–540.

Baltes, P. B., & Lindenberger, U. (1997). Emergence of a powerfulconnection between sensory and cognitive functions across the adult lifespan: A new window to the study of cognitive aging? Psychology andAging, 12, 12–21.

Bennett, D. A., Shannon, K. M., Beckett, L. A., Goetz, C. G., & Wilson,R. S. (1997). Metric properties of nurses’ ratings of parkinsonian signswith a modified Unified Parkinson’s Disease Rating Scale. Neurol-ogy, 49, 1580–1587.

Bennett, D. A., Shannon, K. M., Beckett, L. A., & Wilson, R. S. (1999).

Dimensionality of parkinsonian signs in aging and Alzheimer’s disease.Journal of Gerontology: Medical Sciences, 54A, M191–M196.

Benton, A. L., Sivan, A. B., Hamsher, K. deS., Varney, N. R., & Spreen,O. (1994). Contributions to neuropsychological assessment (2nd ed.).New York: Oxford University Press.

Berg, L., Miller, J. P., Bary, J., Rubin, E. H., Morris, J. C., & Figiel, G.(1992). Mild senile dementia of the Alzheimer type. Annals of Neurol-ogy, 31, 242–249.

Blair, J. R., & Spreen, O. (1989). Predicting premorbid IQ: A revision ofthe National Adult Reading Test. The Clinical Neuropsychologist, 3,129–136.

Brayne, C., Gill, C., Paykel, E. S., Huppert, F., & O’Connor, D. W. (1995).Cognitive decline in an elderly population: A two wave study of change.Psychological Medicine, 25, 673–683.

Burke, D. M., & Peters, L. (1986). Word associations in old age: Evidencefor consistency in semantic encoding in adulthood. Psychology andAging, 1, 283–292.

Christensen, H., MacKinnon, A. J., Korten, A. E., Jorm, A. F., Henderson,A. S., Jacomb, P., & Rodgers, B. (1999). An analysis of diversity in thecognitive performance of elderly community dwellers: Individual dif-ferences in change scores as a function of age. Psychology and Ag-ing, 14, 365–379.

Colsher, P., & Wallace, R. (1991). Longitudinal application of cognitivefunction measures in a defined population of community-dwelling el-ders. Annals of Epidemiology, 1, 215–230.

Cooper, J. A., & Sagar, H. J. (1993). Incidental and intentional recall inParkinson’s disease: An account based on diminished attentional re-sources. Journal of Clinical and Experimental Neuropsychology, 15,713–731.

Cooper, J. A., Sagar, H. J., Doherty, S. M., Jordan, N., Tidswell, P. T., &Sullivan, F. V. (1992). Different effects of dopaminergic and anticho-linergic therapies on cognitive and motor function in Parkinson’s dis-ease: A follow up study of untreated patients. Brain, 115, 1701–1725.

Cooper, J. A., Sagar, H. J., Jordan, N., Harvey, N. S., & Sullivan, E. V.(1991). Cognitive impairment in early, untreated Parkinson’s diseaseand its relationship to motor disability. Brain, 114, 2095–2122.

Craik, F. I. M. (1986). A functional account of age differences in memory.In E. Klix & H. Hagendorf (Eds.), Human memory and cognitivecapabilities: Mechanisms and performances (pp. 409–422). Amster-dam: Elsevier Science.

Cronbach, L. J., & Furby, L. (1970). How we should measure“change”—or should we? Psychological Bulletin, 74, 68–80.

Deeg, D. J. H., Hofman, A., & van Zonneveld, R. J. (1990). The associ-ation between change in cognitive function and longevity in Dutchelderly. American Journal of Epidemiology, 132, 973–982.

Dikmen, S. S., Heaton, R. K., Grant, I., & Temkin, N. R. (1999). Test–retest reliability and practice effects of Expanded Halstead–Reitan Neu-ropsychological Test Battery. Journal of the International Neuropsycho-logical Society, 5, 346–356.

Ekstrom, R. B., French, J. W., Harman, H. H., & Kermen, D. (1976).Manual for kit of factor-referenced cognitive tests. Princeton, NJ: Edu-cational Testing Service.

Emery, C. F., Pedersen, N. L., Svartengren, M., & McClearn, G. E. (1998).Longitudinal and genetic effects in the relationship between pulmonaryfunction and cognitive performance. Journal of Gerontology: Psycho-logical Sciences, 53B, P311–P317.

Evans, D. A., Beckett, L. A., Albert, M. S., Hebert, L. E., Scherr, P. A.,Funkenstein, H. H., & Taylor, J. O. (1993). Level of education andchange in cognitive function in a community population of older per-sons. Annals of Epidemiology, 3, 71–77.

Farmer, M. E., Kittner, S. J., Rae, D. S., Bartko, J. J., & Reiger, D. A.(1995). Education and change in cognitive function: The EpidemiologicCatchment Area Study. Annals of Epidemiology, 5, 1–7.

Feskens, E. J. M., Havekes, L. M., Kalmijn, S., de Knijff, P., Launer, L. J.,


This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

& Kromhout, D. (1994). Apolipoprotein �4 allele and cognitive declinein elderly men. British Medical Journal, 309, 1202–1206.

Folstein, M., Folstein, S., & McHugh, P. (1975). Mini-Mental State: Apractical method for grading the mental state of patients for the clinician.Journal of Psychiatric Research, 12, 189–198.

Fozard, J. L., Vercruyssen, M., Reynolds, S. L., Hancock, P. A., & Quilter,R. E. (1994). Age differences and changes in reaction time: The Balti-more Longitudinal Study of Aging. Journal of Gerontology: Psycholog-ical Sciences, 49, P179–P189.

Giambra, L. M., Arenberg, D., Zonderman, A. B., Kawas, C., & Costa,P. T., Jr. (1995). Adult life span changes in immediate visual memoryand verbal intelligence. Psychology and Aging, 10, 123–139.

Gibbons, R. D., Hedeker, D., Elkin, I., Waternaux, C., Kraemer, H. C.,Greenhouse, J. B., Shea, M. T., Imber, S. D., Sotsky, S. M., & Watkins,J. T. (1993). Some conceptual and statistical issues in analysis oflongitudinal psychiatric data. Archives of General Psychiatry, 50, 739–750.

Gold, D. P., Andres, D., Etezadi, J., Arbuckle, T., Schwartzman, A., &Chaikelson, J. (1995). Structural equation model of intellectual changeand continuity and predictors of intelligence in older men. Psychologyand Aging, 10, 294–303.

Goodglass, H., & Kaplan, D. (1983). The assessment of aphasia andrelated disorders (2nd ed.). Philadelphia: Lea & Febiger.

Grober, E., & Sliwinski, M. (1991). Development and validation of amodel for estimating premorbid verbal intelligence in the elderly. Jour-nal of Clinical and Experimental Neuropsychology, 13, 933–949.

Haxby, J. V., Grady, C. L., Koss, E., Horwitz, B., Heston, L., Schapiro, M.,et al. (1990). Longitudinal study of cerebral metabolic asymmetries andassociated neuropsychological patterns in early dementia of the Alzhei-mer type. Archives of Neurology, 47, 753–760.

Hebert, L. E., Wilson, R. S., Gilley, D. W., Beckett, L. A., Scherr, P. A.,Bennett, D. A., & Evans, D. A. (2000). Decline of language amongwomen and men with Alzheimer’s disease. Journal of Gerontology:Psychological Sciences, 55B, P354–P360.

Hertzog, C., Dixon, R. A., & Hultsch, F. A. (1992). Intraindividual changein text recall of the elderly. Brain and Language, 42, 248–269.

Hopp, G. A., Dixon, R. A., Grut, M., & Backman, L. (1997). Longitudinaland psychometric profiles of two cognitive status tests in very old adults.Journal of Clinical Psychology, 53, 673–686.

Horn, J. L. (1968). Organization of abilities and the development ofintelligence. Psychological Review, 75, 242–259.

Hultsch, D. F., Hertzog, C., Dixon, R. A., & Small, B. J. (1998). Memorychange in the aged. New York: Cambridge University Press.

Izaks, G. J., Gussekloo, J., Dermout, K. M. T., Heeren, T. J., & Ligthart,G. J. (1995). Three-year follow-up of Mini-Mental State Examinationscore in community residents aged 85 and over. Psychological Medi-cine, 25, 841–848.

Jacqmin-Gadda, H., Fabrigoule, C., Commenges, D., & Dartigues, J.-F.(1997). A 5-year longitudinal study of the Mini-Mental State Examina-tion in normal aging. American Journal of Epidemiology, 145, 498–506.

Kaplan, E. F., Goodglass, H., & Weintraub, S. (1983). The Boston NamingTest. Philadelphia: Lea & Febiger.

Kordower, J. H., Chu, Y., Stebbins, G. T., DeKosky, S. T., Cochran, E. J.,Bennett, D. A., & Mufson, E. J. (2001). Loss and atrophy of layer IIentorhinal cortex neurons in elderly people with mild cognitive impair-ment. Annals of Neurology, 49, 202–213.

Laird, N., & Ware, J. (1982). Random-effects models for longitudinal data.Biometrics, 38, 963–974.

Lindenberger, U., & Baltes, P. B. (1994). Sensory functioning and intel-ligence in old age: A strong connection. Psychology and Aging, 9,339–355.

Lyketsos, C. G., Chen, L.-S., & Anthony, J. C. (1999). Cognitive declinein adulthood: An 11.5-year follow-up of the Baltimore EpidemiologicalCatchment Area Study. American Journal of Psychiatry, 156, 58–65.

MacKay, D. G., Abrams, L., & Pedroza, M. J. (1999). Aging on the inputversus output side: Theoretical implications of age-linked asymmetriesbetween detecting versus retrieving orthographic information. Psychol-ogy and Aging, 14, 3–17.

Malec, J. F., Smith, G. E., Ivnik, R. J., Peterson, R. C., & Tangalos, E. G.(1996). Clusters of impaired normal elderly do not decline cognitivelyin 3 to 5 years. Neuropsychology, 10, 66–73.

McCarty, S. M., Siegler, I. C., & Logue, P. E. (1982). Cross-sectional andlongitudinal patterns of three Wechsler Memory Scale subtests. Journalof Gerontology, 37, 169–175.

McKhann, G., Drachman, D., Folstein., M., Katzman, R., Price, D., &Stadlan, E. (1984). Clinical diagnosis of Alzheimer’s disease: Report ofthe NINCDS/ADRDA work group under the auspices of Department ofHealth and Human Services Task Force on Alzheimer’s Disease. Neu-rology, 34, 939–944.

Mitrushina, M., & Satz, P. (1991). Effect of repeated administration of aneuropsychological battery in the elderly. Journal of Clinical Psychol-ogy, 47, 790–801.

Morris, J., Heyman, A., Mohs, R., Hughes, J., van Belle, G., Fillenbaum,G., et al. (1989). The consortium to establish a registry for Alzheimer’sdisease (CERAD): I. Clinical and neuropsychological assessment ofAlzheimer’s disease. Neurology, 39, 1159–1165.

Mufson, E. J., Chen, E.-Y., Cochran, E. J., Beckett, L. A., Bennett, D. A.,& Kordower, J. H. (1999). Entorhinal cortex �-amyloid load in individ-uals with mild cognitive impairment. Experimental Neurology, 158,469–490.

Nelson, H. E., (1982). National Adult Reading Test (NART): Test manual.Windsor, England: NFER Nelson.

Owens, W. A. (1966). Age and mental abilities: A second adult follow-up.Journal of Educational Psychology, 57, 311–325.

Paolo, A. M., Troster, A. I., & Ryan, J. J. (1997). Test–retest stability of theCalifornia Verbal Learning Test in older persons. Neuropsychology, 11,613–616.

Prince, M., Lewis, G., Bird, A., Blizard, R., & Mann, A. (1996). Alongitudinal study of factors predicting change in cognitive test scoresover time, in an older hypertensive population. Psychological Medi-cine, 26, 555–568.

Rabbitt, P. M. A. (1993). Does it all go together when it goes? TheNineteenth Bartlett Memorial Lecture. Quarterly Journal of Experimen-tal Psychology: Human Experimental Psychology, 46, 385–434.

Rand, W. M. (1971). Objective criteria for the evaluation of clusteringmethods. Journal of the American Statistical Association, 66, 846–850.

Raven, J. C., Court, J. H., & Raven J. (1992). Manual for Raven’sprogressive matrices and vocabulary: Standard Progressive Matrices.Oxford, England: Oxford Psychologists Press.

Rockwood, K., Wentzel, C., Hachinski, V., Hogan, D. B., MacKnight, C.,& McDowell, I. (2000). Prevalence and outcomes of vascular cognitiveimpairment. Neurology, 54, 447–451.

Rogosa, D. (1988). Myths about longitudinal research. In K. W. Schaie,R. T. Campbell, & W. Meredith (Eds.), Methodological issues in agingresearch (pp. 171–209). New York: Springer Publishing Company.

Salthouse, T. A. (1988). Initiating the formalization of theories of cognitiveaging. Psychology and Aging, 3, 3–16.

Salthouse, T. A. (1993). Speed and knowledge as determinants of adult agedifferences in verbal tasks. Journal of Gerontology: Psychological Sci-ences, 48, P29–P36.

Salthouse, T. A., Fristoe, N., & Rhee, S. H. (1996). How localized areage-related effects on neuropsychological measures? Neuropsychol-ogy, 10, 272–285.

SAS Institute, Inc. (2000). SAS/STAT User’s Guide, Version 8 [Softwaremanual]. Cary, NC: Author.

Schaie, K. W. (1996). Intellectual development in adulthood: The SeattleLongitudinal Study. New York: Cambridge University Press.

Siegler, I. C., (1983). Psychological aspects of the Duke longitudinal

192 WILSON ET AL.

This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

studies. In K. W. Schaie (Ed.), Longitudinal studies of adult psycholog-ical development (pp. 136–190). New York: Guilford Press.

Siegler, I. C., & Botwinick, J. (1979). A long-term longitudinal study ofintellectual ability of older adults: The matter of selective subject attri-tion. Journal of Gerontology, 34, 242–245.

Small, B. J., Dixon, R. A., Hultsch, D. F., & Hertzog, C. (1999). Longi-tudinal changes in quantitative and qualitative indicators of word andstory recall in young-old and old-old adults. Journal of Gerontology:Psychological Sciences, 54, P107–P115.

Small, S. A., Stern, Y., Tang, M., & Mayeux, R. (1999). Selective declinein memory function among healthy elderly. Neurology, 52, 1392–1396.

Smith, A. (1982). Symbol Digit Modalities Test manual—revised. LosAngeles: Western Psychological Services.

Taylor, J. L., Miller, T. P., & Tinklenberg, J. R. (1992). Correlates ofmemory decline: A 4-year longitudinal study of older adults with mem-ory complaints. Psychology and Aging, 7, 185–193.

Teri, L., Hughes, J., & Larson, E. (1990). Cognitive deterioration inAlzheimer’s disease: Behavioral and health factors. Journal of Geron-tology: Psychological Sciences, 45, 58–63.

Troyer, A., Moscovitch, M., & Winocur, G. (1997). Clustering and switch-ing as two components of verbal fluency: Evidence from younger andolder healthy adults. Neuropsychology, 11, 138–146.

Unger, J. M., van Belle, G., & Heyman, A. (1999). Cross-sectional versuslongitudinal estimates of cognitive change in nondemented older people:A CERAD study. Journal of the American Geriatrics Society, 47,559–563.

Verhaegen, P., & Salthouse, T. A. (1997). Meta-analyses of age–cognitionrelations in adulthood: Estimates of linear and nonlinear age effects andstructural models. Psychological Bulletin, 122, 231–249.

Watson, F. L., Pasteur, M. A. L., Healy, D. T., & Hughes, E. A. (1994).Nine parallel versions of four memory tests: An assessment of formequivalence and the effects of practice on performance. Human Psycho-pharmacology, 9, 51–61.

Wechsler, D. (1987). Wechsler Memory Scale—Revised manual. San An-tonio, TX: Psychological Corporation.

Wilson, R. S., Beckett, L. A., Bennett, D. A., Albert, M. S., & Evans, D. A.

(1999). Change in cognitive function in older persons from a communitypopulation: Relation to age and Alzheimer’s disease. Archives of Neu-rology, 56, 1274–1279.

Wilson, R. S., Bennett, D. A., Gilley, D. W., Beckett, L. A., Barnes, L. L.,& Evans, D. A. (2000). Premorbid reading activity and patterns ofcognitive decline in Alzheimer’s disease. Archives of Neurology, 57,1718–1723.

Wilson, R. S., Bennett, D. A., Gilley, D. W., Beckett, L. A., Schneider,J. A., & Evans, D. A. (2000a). Progression of parkinsonian signs inAlzheimer’s disease. Neurology, 54, 1284–1289.

Wilson, R. S., Bennett, D. A., Gilley, D. W., Beckett, L. A., Schneider,J. A., & Evans, D. A. (2000b). Progression of parkinsonism and loss ofcognitive function in Alzheimer’s disease. Archives of Neurology, 57,855–860.

Wilson, R. S., Gilley, D. W., Bennett, D. A., Beckett, L. A., & Evans, D. A.(2000a). Hallucinations, delusions, and cognitive decline in Alzheimer’sdisease. Journal of Neurology, Neurosurgery and Psychiatry, 69, 172–177.

Wilson, R. S., Gilley, D. W., Bennett, D. A., Beckett, L. A., & Evans, D. A.(2000b). Person-specific paths of cognitive decline in Alzheimer’s dis-ease and their relation to age. Psychology and Aging, 15, 18–28.

Youngjohn, J. R., & Crook, T. H. (1993). Stability of everyday memory inage-associated memory impairment: A longitudinal study. Neuropsy-chology, 7, 406–416.

Zelinski, E. M., & Burnight, K. P. (1997). Sixteen-year longitudinal andtime lag changes in memory and cognition in older adults. Psychologyand Aging, 12, 503–513.

Zhu, L., Fratiglioni, L., Guo Z., Aguero-Torres, H., Winblad, B., &Viitanen, M. (1998). Association of stroke with dementia, cognitiveimpairment, and functional disability in the very old: A population-basedstudy. Stroke, 29, 2094–2099.

Received August 25, 2000Revision received April 30, 2001

Accepted May 11, 2001 �


This

doc

umen

t is c

opyr

ight

ed b

y th

e A

mer

ican

Psy

chol

ogic

al A

ssoc

iatio

n or

one

of i

ts a

llied

pub

lishe

rs.

This

arti

cle

is in

tend

ed so

lely

for t

he p

erso

nal u

se o

f the

indi

vidu

al u

ser a

nd is

not

to b

e di

ssem

inat

ed b

road

ly.

Individual Differences in Rates of Change in Cognitive ... · Individual Differences in Rates of...

Documents

Transcript of Individual Differences in Rates of Change in Cognitive ... · Individual Differences in Rates of...