Age and Ability Affect Practice Gains inLongitudinal Studies of Cognitive Change

Patrick Rabbitt,1 Mary Lunn,2 Danny Wong,2 and Mark Cobain3

1Department of Experimental Psychology, University of Oxford, England, and University of West Australia2Department of Statistics, University of Oxford, England.

3Department of Research and Development, Unilever PLC, London, England.

During a 20-year longitudinal study, 5,842 participants aged 49 to 93 years significantly improved over two to foursuccessive experiences of the Heim AH4-1 intelligence test (first published in 1970), even with between-testintervals of 4 years and longer. After we considered significant attrition by death and dropout and the effects ofgender, socioeconomic advantage, and recruitment cohort, we found that participants with high intelligence testscores showed greater improvement than did those with lower intelligence test scores. Practice gains also reducedwith age, even after we took into consideration the individual differences in intelligence test scores. Thisemphasizes the methodological point that neglect of individual differences in improvement during longitudinalstudies underestimates age-related changes in younger and more able participants and the theoretical point that,like all experiences during everyday life, participation in longitudinal studies alters the ability of aging humans tocope with cognitive demands to different extents according to their baseline abilities.

Key Words: Cognitive gainLongitudinal studyPractice gains.

A MAIN goal for cognitive gerontology is to studyindividual differences in trajectories of cognitive changein old age and so to identify factors that accelerate and retardchanges in mental abilities. This requires longitudinal designsin which participants repeatedly take the same, or very similar,cognitive tests. A new problem then arises because practicegains that are due to repeated testing can disguise declinesassociated with increasing age and frailty. This has long beenrecognized as a theoretical possibility (Schaie, 1965; Schaie,Labouvie, & Barrett, 1973) and has been shown to be a problemin the assessment of testretest reliability of clinical diagnosticmeasures (Beglinger, 2005; Dikmen, Heaton, Grant, & Temkin,1999; Falleti, 2006; Kulik, Chen-Lin, Kulik, & Bangert, 1984;Mitrushima & Satz, 1991). However, it is only recently that thedevelopment of new statistical models has made it possible todemonstrate formally that when testretest intervals are as longas 4 to 7 years, practice gains are large enough to cause seriousunderestimations of the true rates of declines caused by increas-ing age, pathology, and frailty (Ferrer, Salthouse, McArdle,Stewart, & Schwartz, 2004; Ferrer, Salthouse, Stewart, &Schwartz, 2004; Rabbitt, Diggle, Holland, & McInnes, 2004;Rabbitt, Diggle, Smith, Holland, & McInnes, 2001). Because itis clear that practice gains occur in longitudinal studies, a furtherpossibility is that they may vary between individuals. Thisseems probable because many brief laboratory experiments havefound that elderly people learn new material more slowly (forreviews, see Craik & Jennings, 1992; Kausler, 1990; andRabbitt, 2002). Another possibility, also suggested by compar-isons on brief training studies, is that as intervals betweensuccessive testing become very long, older participants willforget more than young participants and so appear to benefit lessfrom repeated testing. In either case, the acceleration of age-related cognitive decline would be overestimated.

There is also evidence that practice gains may vary withindividual differences in general fluid intelligence (gF). Rabbitt,

Bithell, Perdicou, Stollery, and Moore (2007 submitted)practiced 93 volunteers aged from 61 to 82 years on eightdifferent tests of verbal learning, spatial learning, motorlearning, and information processing speed. These researchersfound that initial scores and subsequent rates of improvementwere uncorrelated between different cognitive tests but weresignificantly predicted by unadjusted scores on three differenttests of gF, namely, the Heim 1970 AH4-1 and AH4-2 and theCattell and Cattell (1960) Culture Fair test. Other studies havealso shown that, irrespective of their ages, individuals withhigher intelligence test scores learn new material more rapidlyand remember it better than do those with lower intelligencetest scores (e.g., Rabbitt & Anderson, 2006). This made ituseful for us to check whether there are indeed marked indi-vidual differences in practice gains during longitudinal studiesand whether these gains systematically differ with an indi-viduals age between 60 and 82 years and with an individualsintelligence test scores on entry to a longitudinal study.

Apart from these methodological issues, individual differ-ences in practice gains are in themselves substantively interest-ing as a further index of the age-related cognitive changes thatlongitudinal studies purport to assess. To confidently extrapo-late from age-related changes observed in laboratory studies tochanges in the ability to manage everyday life, we mustrecognize reciprocities between what people do and whathappens to them and, even in old age, how and by how muchtheir experiences alter their abilities.

In order to estimate and compare practice gains accurately, wealso need to take account of some other frequently neglectedfactors. One is self-selection on entry. Volunteers for de-manding longitudinal studies are unusually healthy, able, andhighly motivated members of their age groups (Lachman,Lachman, & Taylor, 1982). This elite bias may be unavoidablebut, because volunteers are typically recruited in successivewaves over many years, it is at least possible to check whether

Journal of Gerontology: PSYCHOLOGICAL SCIENCES Copyright 2008 by The Gerontological Society of America2008, Vol. 63B, No. 4, P235P240

P235

at Galter H

ealth Sciences Library on June 26, 2015http://psychsocgerontology.oxfordjournals.org/

Dow

nloaded from

apparent differences in trajectories of change and in practiceeffects are affected by demographic factors such as gender, age,differences between recruitment cohorts, differences in levelsof socioeconomic advantage, and geographical locations ofresidence. A more difficult problem is to allow for selectiveattrition that occurs because older, frailer, and less ableindividuals die and withdraw earlier than do others. The longerstudies continue, the more elite and atypical of their age groupsthe survivors become (e.g., Lachman, Lachman, & Taylor;Rabbitt, 2002; Rabbitt, Lunn, & Wong, 2005; Rabbitt, Watson,Donlan, Bent, & McInnes, 1994a, 1994b; Rabbitt, Wong, &Lunn, in press; Schaie et al., 1973). Because true trajectories ofchange cannot be estimated unless deaths and dropouts are takeninto consideration, these must also be included in the analysis.

Data collected during the University of Manchester longitu-dinal study of cognitive change in healthy old age, described indetail elsewhere (Rabbitt, Diggle, Holland, McInnes, Bent, et al.,2004), allowed us to examine individual differences in improve-ments that were due to practice during two to four succes-sive experiences of the Heim (1970) AH4-1 intelligence test,administered to participants at 4-year intervals over total periodsof 8 to 16 years, after the effects of recruitment cohort, city ofresidence, gender, and socioeconomic advantage and selectiveattrition by death and dropout had been taken into consideration.

METHODS

Participants, Procedure, and MaterialsResearchers recruited a sample of 5,842 volunteers, that is,

2,615 residents of Greater Manchester and 3,277 residents ofNewcastle-upon-Tyne, United Kingdom, by appeals on localmedia and by word of mouth. The 1,711 men were between theages of 49 and 93 years (M 65.6, SD 7.7), and the 4,131women were between 49 and 92 years (M 64.4, SD7.8) Alltraveled independently to the Department of Psychology ateither the University of Manchester or the University ofNewcastle, where they completed batteries of cognitive tasks inquiet rooms supervised by two experienced testers. Participantswere each reimbursed expenses of 5 (UK) per session. Asearch by HM Registry Office UK identified all 2,342 deathsbetween 1983 and the close of the census on July 31, 2004.Between 1983, when the study began, and July of 2003, when itended, there were 3,204 dropouts, of whom 1,208 also diedbefore the 2004 census. Because many dropouts could only beidentified by failures to return for further testing, dates ofdropout are recorded as the last session attended. The remaining1,996 dropout participants did not drop out before July of 2003and also survived the July 2004 census.

Earlier analyses by some of us and our colleagues, namely,Rabbitt, Diggle, Holland, McInnes, Bent, and colleagues(2004), examined only Newcastle residents and found strongpractice effects on the Heim (1970) AH4-1 intelligence test,contrasting with relatively slight, though still significant, effectson verbal learning tasks and vocabulary tests. As AH4-1 scoresshow the greatest practice gains, they were selected as being themost sensitive indices of possible individual differences inpractice effects. The AH4-1 intelligence test consists of 64problems with equal numbers of logic problems, verbalcomparisons, arithmetic problems, and number series. After

introductory practice on one question from each of these cate-gories, participants answer as many problems as possible within10 minutes. Scores are the percentages of correct answers.The results analyzed are from one Heim (1970) AH4-1 groupintelligence test, which was included in a test battery thatwas repeated at 4-year intervals. Results for Manchester andNewcastle are closely similar and so we combine them.

Previous analyses of data from the Newcastle sample by ourearlier group (Rabbitt, Diggle, Holland, McInnes, Bent, et al.,2004) found that participants levels of performance on cognitivetests markedly vary with their levels of socioeconomic advantage(SEA) as categorized by reference to the UK Office ofPopulation Census and Surveys Classification of OccupationalCategories (1980). Categories are SEA C1 (n261), made up ofprofessionals such as doctors, lawyers, senior managers, andacademics; SEA C2, (n 1,854), made up of other professionalssuch as schoolteachers, pharmacists, and junior managers; SEAC3N (n2,064), made up of skilled nonmanual workers such assecretaries; SEA C3M (n 771), made up of skilled manualworkers such as craftsmen, joiners, fitters, and machinists; SEAC4, (n433), made up of nonskilled nonmanual workers such asclerical assistants and storekeepers; and SEA C5 (n 40), madeup of nonskilled manual workers such as laborers, cleaners, andjanitors. The remaining 427 did not record occupational data.Our group found that cognitive test scores also significantlydiffer with gender: Men scored higher on tests of gF and womenhigher on tests of verbal memory and learning. Recruitmentcohorts also differ significantly in test scores and levels ofsocioeconomic advantage. Accordingly, we also entered occu-pational category and gender into the analyses.

We found that cognitive test scores, and rates of change,markedly varied between subsets of individuals who completedthe study between 1983 and 2003, those who died during thecourse of the study, those who dropped out during the study butsurvived beyond a census of deaths completed by HM Registryof Births, Marriages and Deaths, Stockport UK in July of 2004,and those who dropped out during the study but subsequentlyalso died before the 2004 census (Rabbitt, Lunn, & Wong,2005). As the incidence and timing of deaths and dropoutsmight affect the sizes of practice effects, we found necessary totest for interactions between practice and deathdropout status.Accordingly, we divided participants into 11 subgroupsaccording to their histories of survival, dropout, or dropoutfollowed by death with respect to the time points of the fourquadrennial test sessions at which the Heim (1970) AH4-1 testwas administered. These groups were as follows.

Group C completed the study and survived the 2004 censusof deaths (n 1,510); Group D1 died between Test Session 1and Test Session 2 (n 365); Group D2 died between TestSession 2 and Test Session 3 (n409); Group D3 died betweenTest Session 3 and Test Session 4 (n 246); Group D4 diedbetween Test Session 4 and the 2004 census of deaths (n116);Group WD1 withdrew before Test Session 2 and subsequentlydied (n745); Group WD2 withdrew before Test Session 3 andsubsequently died (n 354); Group WD3 withdrew before TestSession 4 and subsequently died (n109); Group W1 withdrewbefore Test Session 2 and survived beyond the 2004 census (n1,013); Group W2 withdrew before Test Session 3 and survivedbeyond the 2004 census (n 595); and Group W3 withdrewbefore Test Session 4 and survived the 2004 census (n 388).

RABBITT ET AL.P236

at Galter H

ealth Sciences Library on June 26, 2015http://psychsocgerontology.oxfordjournals.org/

Dow

nloaded from

Because our study aim was to examine, independently, theeffects of age and of general intellectual ability on practiceeffects, we found it necessary to have a different measure ofability than scores on the AH4-1, for which practice data wereanalyzed. This was available because the Heim (1970) AH4-2,a nonverbal test of general fluid ability, had been administeredto all participants on each testing occasion cotemporaneouslywith the AH4-1. Within the entire sample, the correlationbetween age-unadjusted percentage correct of AH4-1 and AH4-2scores is r .84, so AH4-2 scores are a good independentmeasure of gF and, specifically, of individual differences inability on the AH4-1. The AH4-2 intelligence test also consistsof 64 problems. Each consists of a series of five line drawingswith a missing element that must be supplied from among fiveprovided alternatives. Solutions of series require mental rota-tion, addition, and subtraction or irregular shapes and recogni-tion of logical progressions. After introductory practice on onequestion from each of these categories, participants attempt tosolve as many problems as possible within 10 minutes. Scoresused are percentages of correct answers.

We conducted our analysis by using a linear mixed effectsmodel with fixed effects including age, demographic factorssuch as gender, socioeconomic advantage, city of residence, andrecruitment cohort, AH4-2 score, practice session, and death andwithdrawal group. The random effects were an intercept for eachindividual and an individual age effect, so that variance betweenindividuals increased with age. A key point of the analysis wasthe introduction of the group effect, allowing the groups from Cto W3 to influence the regression coefficients of the usual fixedeffects. This is a so-called pattern-mixture model. If theregression on the groups, which could include interaction terms,is significant, then this model allows the time and type ofdropout to influence the effects of other factors (Little, 1993).

RESULTS

Effects of Age on Improvement With PracticeThe results of a linear mixed effects pattern-mixture model

are shown in Table 1. Age is centered at 70 years, because thisis close to the mean age of the sample. The linear effect of ageis highly significant and the significant quadratic age termindicates that rates of decline in participantsAH4-1 scoresaccelerate as they grow older. The significant difference forgender occurs because average scores for men are 1.57% higherthan average scores for women. There are also highlysignificant effects of socioeconomic advantage and of re-cruitment year. There are highly significant differences betweengroups with different histories of survival, death, and dropoutover successive time points during the study. Finally, aftervariance associated with all of these other factors has beentaken into consideration, practice gains are robustly significant.Participants score significantly higher on their second, third, orfourth experience than they do on their first experience of theAH4-1 test. The key finding of interest is the significantinteraction between the effects of age and the amount ofpractice gains between Sessions 1 and 2, Sessions 1 and 3, andSessions 1 and 4. These occur because, on all these transitions,the older participants benefit less from practice than theyounger participants do. Another new finding is the Age 3

Gender interaction that occurs because men experience morerapid age-related declines than do women. No other interactionsare significant, so there is no evidence that the amount ofpractice gain is affected by proximity to death or to dropout.

Effects of Intelligence on Improvement With PracticeTo examine the effects of individual differences in gF (AH4-2

scores) on practice gains, an obvious procedure was to fita linear term in AH4-2 and examine the interactions. However,in the present case this was clearly inappropriate, and the fit of

Table 1. Estimated Parameters for the Mean AH4-1

Percentage Score

Parameter Estimate SE t p

Intercept 53.04 1.08 48.24 ,.0001Age 70 0.76 0.06 12.83 ,.0001(Age 70)2 0.01 0.004 2.50 .0126Practice effect

Practice 2 vs practice 1 3.94 0.31 12.81 ,.0001Practice 3 6.45 0.53 12.25 ,.0001Practice 4 8.37 0.92 9.12 ,.0001

Gender

Female vs male 1.57 0.65 2.42 .0157Socioeconomic status

C1 vs C3(N) 9.05 1.28 7.37 ,.0001C2 6.58 0.70 9.37 ,.0001C3(M) 7.96 0.96 8.26 ,.0001C4 11.33 1.24 9.17 ,.0001C5 17.07 3.45 4.95 ,.0001Missing 2.47 1.16 2.12 .0337

Entry year

1986 vs 1985 0.68 0.71 0.96 .33841987 1.83 1.13 1.62 .10561988 3.00 1.00 3.02 .00261990 9.27 2.84 3.26 .0011

1991 0.98 1.45 0.67 .5001

1992 3.59 0.96 3.73 .0002

Pattern

D1 vs C 8.91 1.71 5.22 ,.0001D2 5.70 1.33 4.29 ,.0001D3 2.87 1.67 1.72 .0864D4 3.14 3.47 0.91 .3646WD1 9.71 1.02 9.47 ,.0001WD2 7.69 1.53 5.02 ,.0001WD3 6.20 3.91 1.59 .1128W1 9.37 0.85 11.08 ,.0001W2 4.35 1.00 4.34 ,.0001W3 2.06 1.23 1.68 .0937

Interactions

(Age 70) 3 Practice 2 0.15 0.05 3.01 .0026(Age 70) 3 Practice 3 0.32 0.09 3.57 .0004(Age 70) 3 Practice 4 0.40 0.13 3.05 .0023(Age 70) 3 Gender 0.10 0.04 2.27 .0234

Note: The estimated parameters use a pattern-mixture model with an

11-level pattern of groups experiencing survival, dropout, death, and dropout

followed by death. Note that age is centered at 70 years, which is approxi-

mately the mean age for the sample. AH4-1 Heim test of general fluid intel-ligence. Covariance and residual parameters are as follows: rA, 13.33; rB,0.39; qAB, 0.15; re, 5.54. For the socioeconomic status information, the cate-gory descriptions for levels of socioeconomic advantage are given in the text,

in the Participants, Procedure, and Materials section. For the pattern informa-

tion, the 11 subgroups are also described in that section.

AGE, ABILITY, AND LONGITUDINAL PRACTICE GAINS P237

at Galter H

ealth Sciences Library on June 26, 2015http://psychsocgerontology.oxfordjournals.org/

Dow

nloaded from

this model was not good because there is little practice effect atthe lowest and highest scoring individuals. We also considereda linearquadratic term in AH4-2. Again the fit to the data wasnot good. At this stage it became apparent that (a) therelationship is complicated and (b) the precise form of thiscomplex relationship is of substantive interest in interpreting thepattern of individual differences in practice effects. The form ofthe interaction reveals that estimates of true practice effectsmay be miscomputed both because of ceiling effects for themost able and floor effects for the least able. In other words,the relationship between practice gains and AH4-2 scores isrepresented by an inverted U function, and this feature of thedata is best revealed by subdivision into three groups. Givena requirement for simplicity of interpretation, we should look fortwo change points to give lowest, middling, and highest scoregroups. This selected model gave a good fit to the data, and theinteraction was significant. This procedure would clearly havebeen inappropriate had it been intended be used as a predictionfor individual trajectories. This is not the case. It is only intendedto test for overall differences between ability groups. The resultsof fitting this final model are shown in Table 2.

Because of the strong correlation between AH4-1 and AH4-2scores, we expected the overall effects of level of ability to behighly significant. After we took level of ability intoconsideration, we found that the linear but not the quadraticeffects of age (centered at the average age of 70 years) and theeffects of socioeconomic advantage, recruitment year, andsurvival, death, and withdrawal history were similar to thosefound in the first analysis. An interesting further detail is thatthe effect of gender was now abolished. We interpret this asevidence that all of the variance in AH4-1 scores that isassociated with differences between men and women was nowpicked up by scores on another, different, well-validatednonverbal test of gFthe AH4-2. In other words, differencesin AH4-1 test scores between men and women are not due toany factor, for which gender is a proxy, other than differencesin the particular kind of general fluid mental ability that ismeasured by the AH4-2 and AH4-1 tests. The significant Age3Gender interaction replicates that found in the first analysis,with women showing less age-related decline in AH4-1 scores.The Age 3 Practice interactions found in the first analysisstill remain significant, even after the effects of differences inAH4-2 scores have been taken into consideration. In otherwords, greater age significantly reduces practice gains, evenafter individual differences in general fluid mental ability havebeen taken into account. We interpret this as new evidence thatnot all of the age-related cognitive changes that lead to declinesin efficiency of learning (practice effects) can be explained bydifferences in gF.

The significant interaction between level of ability andpractice gains is a new finding, but the complex nature of thisrelationship requires interpretation. For both the low-abilitygroup and the high-ability group, there are negative interactionswith practice. That is, both the low-ability and the high-abilitygroups benefit less from practice than does the medium-abilitygroup. For the low-ability group this interaction is significantfor comparisons between Session 1 and each of Sessions 2, 3,and 4. For the high-ability group the interaction is significantfor the comparison between Session 1 and Session 2 but not forthe comparisons between Sessions 1 and 3 or Sessions 1 and 4.

Table 2. Estimated Parameters for the Mean AH4-1

Percentage Score: Final Model

Parameter Estimate SE t p

Intercept 52.68 0.99 53.07 ,.0001Age 70 0.52 0.05 10.37 ,.0001Low AH4-2 vs middle AH4-2 21.67 1.17 18.58 ,.0001High AH4-2 20.92 3.16 6.61 ,.0001

Practice effect

Practice 2 vs practice 1 3.11 0.31 10.12 ,.0001Practice 3 5.14 0.52 10.12 ,.0001Practice 4 7.04 0.84 8.35 ,.0001

Gender

Female vs male 0.51 0.60 0.85 .3981Socioeconomic status

C1 vs C3(N) 7.64 1.18 6.47 ,.0001C2 6.29 0.64 9.77 ,.0001C3(M) 6.16 0.89 6.94 ,.0001C4 8.03 1.15 7.00 ,.0001C5 9.68 3.20 3.02 .0025Missing 0.19 1.07 0.17 .8626

Entry year

1986 vs 1985 0.95 0.65 1.45 .14601987 1.69 1.04 1.63 .10341988 2.95 0.91 3.24 .00121990 7.08 2.61 2.71 .0067

1991 0.25 1.33 0.18 .85411992 2.09 0.89 2.36 .0184

Pattern

D1 vs C 7.36 1.57 4.68 ,.0001D2 5.22 1.22 4.29 ,.0001D3 2.05 1.53 1.34 .1800D4 3.03 3.16 0.96 .3369WD1 8.66 0.94 9.21 ,.0001WD2 6.57 1.41 4.67 ,.0001WD3 6.69 3.57 1.87 .0614W1 8.12 0.78 10.44 ,.0001W2 3.92 0.92 4.27 ,.0001W3 2.01 1.12 1.79 .0735

Interactions

(Age 70) 3 Practice 2 0.25 0.03 7.59 ,.0001(Age 70) 3 Practice 3 0.53 0.05 11.44 ,.0001(Age 70) 3 Practice 4 0.69 0.07 9.53 ,.0001(Age 70) 3 Gender 0.11 0.04 2.52 .0117(Age 70) 3 Low AH4-2 0.40 0.14 2.88 .0040(Age 70) 3 High AH4-2 0.04 0.26 0.14 .8855Low AH4-2 3 Practice 2 3.93 1.42 2.77 .0057Low AH4-2 3 Practice 3 5.30 2.72 1.95 .0516Low AH4-2 3 Practice 4 7.66 4.04 1.89 .0582High AH4-2 3 Practice 2 3.72 1.79 2.08 .0377High AH4-2 3 Practice 3 4.40 3.05 1.44 .1500High AH4-2 3 Practice 4 4.30 4.39 0.98 .3272Note: The estimated parameters use a pattern-mixture model with an 11-

level pattern of groups experiencing survival, dropout, death, and dropout fol-

lowed by death. Three levels of AH4-2 scores (AH4-1 and AH4-2 are Heim

tests of general fluid intelligence) are also coded. Note that age is centered at

70 years, which is, approximately, the mean for this sample. Covariance and

residual parameters are as follows: rA, 12.08; rB, 0.37; qAB, 0.16; re, 5.56.For the socioeconomic status information, the category descriptions for levels

of socioeconomic advantage are given in the text, in the Participants, Proce-

dure, and Materials section. For the pattern information, the 11 subgroups are

also described in that section.

RABBITT ET AL.P238

at Galter H

ealth Sciences Library on June 26, 2015http://psychsocgerontology.oxfordjournals.org/

Dow

nloaded from

There is also a significant interaction between age and lowlevel of ability relative to age and either medium or high ability.The age-related decline of 0.52 per year of age overstatesthe amount of decline for those individuals at the bottom levelof the ability range. It is possible that it is the inclusion ofthis interaction that results in the quadratic effect in which agebecomes no longer significant in this model. As in the firstanalysis described earlier, the absence of any interactionbetween practice effects and the occurrence or timing of deathsor withdrawal means that there is no evidence that approach todeath or dropout affects the amount of practice gains afterdifferences in ability have been taken into consideration.

DISCUSSIONThe present analyses of data from the combined Manchester

and Newcastle samples replicate the main findings of earlieranalyses of data from the Newcastle longitudinal samples byRabbitt, Diggle, Holland, McInnes, Bent, and colleagues(2004) and the Manchester and Newcastle sample by Rabbitt,Lunn, and Wong (2005). There are marked effects of socio-economic advantage and of year of entry to the study. Menscore higher than women but women experience less rapidage-related decline. This is of interest because it probablyreflects the fact that, in Western industrialized societies, womenlive longer than men and so experience slower biologicalchanges and retain their competence to a later calendar age, asdistinct from biological age. A further new detail in explorationof gender differences is that differences between men andwomen on one test of gF, the AH4-1, disappear when scoreson a different, highly correlated test of gF, that is, the AH4-2,are taken into consideration. Thus we have no evidence thatfactors other than levels of gF are responsible for gender dif-ferences in performance of intelligence tests.

The significant effects of survival, death, and dropout historyreplicate earlier effects reported by Rabbitt, Lunn, and Wong(2005) and Rabbitt, Lunn, and Wong (2007, submitted).Performance declines with approach to either death or dropoutand the amounts of declines preceding death and dropout areclosely similar; comparing equivalent time points of dropout,we see that the effects of approaching dropout are greater whendropout is shortly followed by death than when dropout isfollowed by survival. This empirically shows that not only theeffects of age-related decline but also the effects of other factorssuch as health or socioeconomic advantage, gender differences,and practiceall of which influence rates of changes incognitive performance over timeare miscalculated unlessboth the occurrence and timing of both deaths and dropouts arealso logged and taken into consideration.

The main point of the present analyses is that, even after wetake into account the effects of initial selection and selectiveattrition of a sample, the marked practice effects duringa prolonged longitudinal study found by some of us in earlierresearch (Rabbitt, Diggle, Holland, & McInnes, 2004) arereplicated on a different and very much larger sample ofparticipants (i.e., the Manchester and Newcastle group ratherthan just the Newcastle cohort of the University of Manchesterlongitudinal study). As in our earlier analyses (Rabbitt, Diggle,Holland, & McInnes, 2004), significant practice improvementsare found even when intervals between successive presentations

of the task are as long as 4 years. Indeed, the present analysesshow that even the oldest participants show gains withquadrennial experiences of the AH4-1 test over periods of 8and of 12 years.

The new questions asked by these analyses were whetherpractice gains during a longitudinal study differ betweenindividuals of different ages and different levels of ability. Thefirst analysis shows that younger participants gain more frompractice than do relatively older participants. The secondanalysis shows that, even after variance associated with agedifferences has been taken into account, higher scores on onetest of general intellectual ability, the AH4-2 intelligence test,are associated with significant increments in practice gains onanother, the AH4-1. The further finding that age differences inpractice remain significant even after effects of differences inunadjusted intelligence test scores have been considered makesthe additional new point that the reduction in practice effectswith increasing age cannot entirely be attributed to the fact thatpeoples levels of gF decline as they grow older. This isinconsistent with recent speculations that age-related differ-ences in all cognitive skills can be parsimoniously treated asdifferences in gF (e.g., Anderson, 1992, Deary 2000). Theparticular age-related cognitive changes that reduce practicegains in this longitudinal study are not entirely captured byscores on a well-validated test of gF.

Because the form of the interaction between AH4-2 test scoregroup and practice gains is complex, it requires interpretation.Both the high-ability and the low-ability groups show lesspractice gains than does the medium-ability group. Manyparticipants in the high-ability group scored the maximumpossible, and many others had such high scores that there waslittle scope for improvement. Thus ceiling effects offera sufficient explanation as to why high-ability participants showless improvement. Within the range of scores in which ceilingeffects are no longer possible so that practice gains can appear,the significant advantage for the middle-ability group over thelow-ability group shows that practice gains do increase with gF.

We must conclude that all previous analyses of longitudinaldata in which only mean values for practice effects have beencalculated have, because of this, significantly underestimatedthe true amounts of age-related declines for younger and moreable participants. For older and less able participants, who showsignificantly less improvement with practice, estimated rates ofage-related declines have been closer to their true values.This finding highlights a nontrivial issue because, unless it isimplemented in further analyses, we cannot confidently addresssome theoretically interesting and socially important questionssuch as whether more and less able individuals experiencedifferent rates of cognitive decline and whether data showgeneration cohort effects. For example, do young participants inlongitudinal studies, who have benefited from recent improve-ments in socioeconomic conditions and in medical care,experience less rapid cognitive decline than earlier generationswho have, historically, been disadvantaged in these respects?To find reliable answers to these questions, we must measureand consider individual differences in practice effects.

Perhaps the most general point that these analyses make isthat individuals cognitive abilities are altered by theirinvolvement in prolonged longitudinal studies, just as by theirinvolvements in all other aspects of their everyday lives. The

AGE, ABILITY, AND LONGITUDINAL PRACTICE GAINS P239

at Galter H

ealth Sciences Library on June 26, 2015http://psychsocgerontology.oxfordjournals.org/

Dow

nloaded from

variety and the particular nature of our experiences significantlyalter our abilities to cope with demands made by our everydaylives or, indeed, by psychological tests. People do not simplyand inexorably decline in mental ability as they grow old. Theireveryday experiences may also degrade or enhance their abilityto cope with the demands that their lives make upon them. Wemay speculate that, as well as learning to cope with particularlife demands and situations, people also gradually learn to adaptto and cope with the changes in their mental abilities broughtabout by neurophysiological aging. It is striking that, even inelderly individuals, very brief episodes of practice, even as briefas 10 minutes on first encounter with the simple problems inthe AH4-1, can bring about domain-specific improvementsthat last for 4 to 7 years (see Rabbitt, Diggle, Holland, McInnes,Bent, et al., 2004). It therefore seems less interesting to con-tinue to explore changes in individuals scores on particularcognitive tests than to study interactions between their baselinelevels of ability and their life experiences, and so the levelsof competence they can achieve and so also the extent to whichthey can maintain, in old age, their ability to cope with theirlives and with any laboratory experiments into which they maybe inveigled.

REFERENCES

Anderson, M. (1992). Intelligence and development: A cognitive theory.Oxford. Blackwell.

Beglinger, A. ( 2005). Practice effects and the use of alternate forms inserial neuropsychological testing. Archives of Clinical Neuropsychol-ogy, 20, 512529.

Craik, F. I. M., & Jennings, J. M. (1992). Human memory. In F. I. M. Craik &T. Salthouse (Eds.), The handbook of aging and cognition (pp. 51110).Hillsdale, NJ: Erlbaum.

Deary, I. J. (2000). Looking down on human intelligence. Oxford, England:Oxford University Press.

Diggle, P. J., Liang, K. Y., & Zeger, S. L. (1994). Analysis of longitudinaldata. Oxford, England: Oxford University Press.

Dikmen, S. S., Heaton, R. K., Grant, J., & Temkin, N. R. (1999). Testretest reliability and practice effects of Expanded HalsteadReitanNeuropsychological Test Battery. Journal of the International Neuro-psychological Society, 5, 346356.

Falleti, K. (2006). Practice effects associated with the repeated assessmentof cognitive function using the Cog State Battery at 10 min, one week,and one month testre-test intervals. Neuropsychology, Developmentand Cognition: Journal of Clinical and Experimental Neuropsychology,28A, 10951112.

Ferrer, E., Salthouse, T., McArdle, J., Stewart, W. F., & Schwartz, B.(2005). Estimating retest effects in longitudinal assessments of cogni-tive functioning in adults between 18 and 60 years of age. Develop-mental Psychology, 40, 813822.

Ferrer, E., Salthouse, T., Stewart, W., & Schwartz, B. (2004). Modeling ageand re-test processes in longitudinal studies of cognitive abilities.Psychology and Aging, 19, 243259.

Heim, A. W. (1970). The AH4 group test of general intelligence. Windsor,England: NFER-Nelson.

Lachman, R., Lachman, J. L., & Taylor, D. W. (1982). Reallocation ofmental resources over the productive lifespan: Assumptions and task

analyses. In F. I. M. Craik & S. Trehub (Eds.), Aging and the cognitiveprocess (pp. 304350). New York: Plenum Press.

Kausler, D. H. (1990). Experimental psychology, cognition and humanaging. New York: Springer-Verlag.

Kulik, J. A, Chen-Lin, C., Kulik, R. L., & Bangert, L. (1984). Effects of

practice on aptitude and achievement test scores. American EducationalResearch Journal, 21, 435447.

Little, R. J. A. (1993). Pattern-mixture models for multi-variate incomplete

data. Journal of the American Statistical Association, 88, 125134.Little, T. D., Lindenberger, U., & Maier, H. (2000). Selectivity and

generalizability in longitudinal research: On the effects of continuers

and dropouts. In T. D. Little, K. U. Schanabel, & J. Baumert, (Eds.),

Modelling longitudinal and multilevel data: Practical issues, appliedapproaches and specific examples (pp. 187200). Mahwah, NJ:Erlbaum.

Mitrushima, M., & Satz, P. (1991). Effect of repeated administration of

a neuropsychological test battery in the elderly. Journal of ClinicalPsychology, 47, 790801.

Nagi, M. H., & Stockwell, E. G. (1973). Socioeconomic differentials in

mortality by cause of death. Health Services Reports, 88, 449456.Office of Population Censuses and Surveys. (1980). Classification of

occupations 1980. London, England: Her Majestys Stationary Office.Rabbitt, P. (2002). Cognitive gerontology. In H. Pashler & J. Wixted (Eds.),

Stevens handbook of experimental psychology (Vol 4., pp. 401453).New York: Wiley.

Rabbitt, P., Diggle, P., Holland, F., & McInnes, L. (2004). Practice and

drop-out effects during a 17-year longitudinal study of cognitive aging.

Journal of Gerontology: Psychological Sciences, 59B, P84P97.Rabbitt, P., Diggle, P., Smith, D., Holland, F., & McInnes, L. (2001).

Identifying and separating the effects of practice and of cognitiveageing during a large longitudinal study of elderly community residents.

Neuropsychologia, 39, 532543.Rabbitt, P., Lunn, M., & Wong, D. (2005). Neglect of dropout

underestimates effects of death in longitudinal studies. Journal ofGerontology: Psychological Sciences, 60B, P106P109.

Rabbitt, P., Watson, P., Donlan, C., Bent, N., & McInnes, L. (1994a).

Subject attrition in a longitudinal study of cognitive performance in

community resident elderly people. In B. J. Vellas, J. L. Albarende, &P. J. Garry (Eds.), Facts and research in gerontology: Epidemiologyand aging (pp. 2934). Paris: Serdi.

Rabbitt, P. M., Watson, P., Donlan, C., Bent, L., & McInnes, L. (1994b).

Subject attrition in a longitudinal study of cognitive performance incommunity-based elderly people. Facts and Research in Gerontology,14, 203207.

Rabbitt, P. M. A., & Anderson, M. (2006). The lacunae of loss? Aging and

the differentiation of cognitive abilities. In E. Bialystok & F. I. M. Craik

(Eds.), Lifespan cognition: Mechanisms of change (pp. 331343).Rabbitt, P. M. A., Diggle, P., Holland, F., McInnes, L., Bent, N., Abson,

V., et al. (2004). The University of Manchester longitudinal study of

cognition in normal healthy old age, 1983 through 2003. Aging,Neuropsychology and Cognition, 11, 245279.

Schaie, K. W. (1965). A general model for the study of developmentalproblems. Psychological Bulletin, 64, 92107.

Schaie, K. W., Labouvie, G. V., & Barrett, T. J. (1973). Selective attritioneffects in a fourteen-year study of adult intelligence. Gerontology, 28,328334.

Received November 30, 2005Accepted January 2, 2008Decision Editor: Thomas M. Hess, PhD

RABBITT ET AL.P240

at Galter H

ealth Sciences Library on June 26, 2015http://psychsocgerontology.oxfordjournals.org/

Dow

nloaded from