Joumal of ExperimenlalPsychology:
l€arning, Memory, and Cognition
1 9 8 9v. o l . 1 5 .N o . 3 . 5 0 7 - 5 1 6
Copyright 1989by the American PsychologicalAsociation, Inc.
Effectsof Adult Age and Working Memory
on Reasoningand SpatialAbilities
Timothy A. Salthouse,DeboraR. D. Mitchell, Eric Skovronek,and ReneeL. Babcock
Georgia Institute of Technology
Three predictionswere derivedfrom the hypothesisthat adult agedifferencesin certain measures
of cognitive functioning are attributable to age-related.reductions
in a processingresourcesuch
as working-memory capacity.Each prediction receivedat leastsomedegreeof empirical support
in a study involving 120 males ranging between 20 and 79 years of age. First, older adults
exhibited greater impairments of performance than did young adults when task complexity
increasedand more demands were placed on the limited processingresources;second, the
magnitudesofthese complexity effectswerehighly conelaledacrossverbal(reasoning)and spatial
(paper folding) tasks. Finally, statistical control of an index of a working-memory processing
nesourceattenuatedthe effectsof ageon the measuresofcognitive perfonnance.It wasconcluded
that further progressin understandingthe mechanismsofthe relation betweenageand cognitive
functioning will require improved conceptualizationsof the nature of working memory or other
hypothesizedmediating constructs.
One of the most frequently mentioned "interpretations" of
adult agedifferencesin fluid measures(cf. Cattell, 197l; Horn,
1982)or Type A (Hebb, 1942)measuresof cognitivefunctioning attributes them to an age-relatedreduction in some
type of general-purposeprocessingresources(seeSalthouse,
1988a,I 988b,for reviewand discussion).
In its simplestform,
the resource argument consists of two assumptions and a
conclusion: (a) the assumption that processingresourcesare
required for many, but not all, cognitive processes;(b) the
assumptionthat, for largelyunspecifiedreasons,increasedage
in adulthood is associatedwith a diminished supply of available processingresources;and (c) the conclusionthat the agerelated reduction in the quantity of processingresourcesresults in poorer performance in tasks containing resourcedemanding processes.Unfortunately, although the frequent
referenceto processingresourcesin discussionsof cognitive
aging phenomena suggeststhat many researchersfind this
generalargument compelling, the processing-resources
interpretation is severelyweakenedby two conceptualproblemsvaguenessof the fundam€ntal construct and ambiguity of the
The frrst problem is that the specific nature of the key
concept in this category of interpretation-processing resources-has seldombeendiscussed.Instead,researchers
employed a variety of synonyms such as elfort, energy, or
capacity without ever specifying exactly what is meant by
these terms. Adjectives are occasionallyadded with the ap
parent intent of increasing the precision of the terms, for
example,cognitiveelfort, mental energy,attentional capacity,
and working-memory capacity, but these elaborations have
typically not beenaccompaniedby more explicit descriptions
This researchwas supportedby National Institute on Aging ResearchGrants KO4AGOO372and ROlAG06826 to the seniorauthor.
Correspondenceconcerning this article should be addressedto
Timothy A. Salthouse,School of Psychology,Georgia Institute of
Technology,Atlanta, Georgia 30332.
that would remove the ambiguity inherent in the resources
A secondmajor problem with past usagesof the conceptof
processingresourcesis that there hasbeenlittle or no attempt
to specify the mechanisms by which processingresources
might influencecognitiveperformance.The primary question
in this context, which hasbeen ignoredin almost all previous
referencesto the resourcesconstructwithin the cognitiveagrng
literature, is how a limited supply of some entity contributes
to lower performance in tasks of memory, reasoning,and
spatial abilities. Furthermore, unless there is at least some
empirical evidencesupporting the hypothesizedcausal relation, it is diflicult to view speculationscontaining references
to processingresourcesas seriousscientifrchypotheses.
The goal of the presentresearchwas to addressthe aforementioned problems while investigatingthe hypothesisthat
at leastsome of the adult agedifferencesin certain cognitive
tasks are mediated by age-relateddeclinesin a type of processingresource.In the following sectionthe reasoningunderlying predictions derived from the resource perspectiveis
discussed;the criteria used in selectingtasks to test those
predictionsare describedin subsequentsections.
Predictions From the Processing-Resource Perspective
Two initial predictions can be derived from the resources
perspectivebasedon the age-complexityphenomenon, that
is, the tendency for the magnitude of the age differencesin
performanceto increasewith the hypothesizedcomplexity of
the task (seeSalthouse,1985,pp. 183-190,for review and
discussion).The first prediction is that qualitatively similar
age-complexitypatterns should be evident across different
cognitive tasks; the second is that the magnitude of the
complexity effectsin different tasksshould be comparablefor
a given individual. The key to both predictionsis the assumption that the performancedecline associatedwith an increase
in task complexity is at least partially attributable to greater
demandson a finite quantity of processingresources.If older
adults have smaller supplies of the relevant processingresourcesthan do young adults, then they would be expected
to suffergreaterperformanceimpairments as the demandson
their more limited resourcesincrease.Moreover, if a common
processingresourceis involved in the complexity-relatedperformance decrementsin different cognitive tasks, then the
quantity ofresourcesavailableto an individual should influence the magnitude of the complexity effects in each task,
and consequently,those effectsshould be significantly correlated acrosstasks.
Another expectationfrom the view that a diminished supply ofgeneral-purposeprocessingresourcescontributesto the
age differencesin different cognitive tasks is that statistical
control ofan index ofthe hypothesizedprocessingresources
should attenuate the magnitude of the age differences in
measuresof cognitiveperformance.The degreeof attenuation
will naturally vary acrossmeasures,dependingon the importanceofthe relevantprocessingresourceto the agedifferences
in particular tasks. Moreover, it is probably unrealistic ever
to expectcompleteelimination of agediflerencesby statistical
control ofan index ofprocessing resourcesbecausethere are
likely to be some nonresourcedeterminants of performance
(e.g., sensoryacuity, quantity or quality of relevant knowledge,etc.) in most cognitive tasks. Nevertheless,if it is true
that agedifferencesin measuresof cognitive performanceare
at least partially determined by age differencesin some type
of processingresource,then statisticalcontrol of an index of
the latter should result in the attenuation ofthe effectsofage
on the former.
Selection of a Resource Index
Examination of the third prediction discussedabove obviously requiresthe availability of an index of the individual's
quantity of processingresources.Two major criteria were
employed in the current project to guide in the selectionof
this index: demonstrablereliability and clear theoreticalrelevance to the resourcesconstruct. Reliability is essentialbecausethe measuremust be stable and consistent if it is to
reflectan enduringtraitlike characteristicsuchas the quantity
of one's processingresources.Someindication of the reliability of the measure is also necessaryto allow meaningful
interpretation ofthe correlational resultsbecausethe rangeof
possible correlations between one variable and another is
obviously limited by the magnitude of the correlations the
variableshave with themselves,that is, by their reliabilities.
There are two aspectsto the theoreticalrelevancecriterion.
The fint is that processingresourcesshould be positively
relatedto various measuresof cognitiveperformance,and the
secondis that they should be negativelyrelated to adult age.
Statedsomewhatdifferently, if processingresourcestruly mediate some of the age-relateddeclines observedin cognitive
functioning, then the index of the quantity of processing
resourcesshould itselfdeclineacrossthe adult years,and there
should be reason to believe that higher amounts of the resourceare associatedwith better levelsof cognitive performance.
A variety ofpotential measuresappearto possess
the desired
characteristics,but determination of the mechanisms by
which processingresourcesinfluence cognitive performance
was consideredmost feasiblewith measuresof the workingmemory conceptualizationof processingresources.Furthermore, working memory is often postulatedto play a central
role in a varietyof cognitivetasks(e.g.,Baddeley1986:Case,
1985),and there have been many reports that aging is associatedwith an impairment in working memory (e.g..Craik &
Rabinowitz, 1984) or active short-term memory (e.g.. Welford. 1958).
Baddeley (1986), Baddeley, Logie, and Nimmo-Smith
(1985),Case(1985),Danemanand Carpenter(1980).and
others have all claimed that a critical aspect of working
memory is that it involves the simultaneousstorageand
processingof information. An important feature of a task
working memory, therefore,is that it must require
the maintenanceof some information during the processing
of that or other information.
The task selectedto assessthis type of working memory is
a modification of the computational span task used by Salthouse(1988a)and Salthouseand Prill (1987).It is similar to
the readingspan task employed by Daneman and Carpenter
( 1980),in which the subjectreadsor listensto sentences
rememberingthe last word in eachsentence,and the counting
span task employed by Case,Kurland, and Goldberg ( 1982),
in which the subjectviewsslidescontaininga variablenumber
of dots, and the number of dots in eachslide must tre counted
while remembering the numbers from earlier slides. The
computational span task consistsof the subject'sperforming
simple arithmetic problems while simultaneouslyremembering a designateddigit from each problem. Earlier research
(Salthouse,1988a)has indicatedthat althoughthe computational spandid not haveimpressivereliability(i.e.,estimated
reliabilitycoefficientsof .65 for 20 young adultsand .57 for
20 older adults), it was neverthelessmoderately correlated
with other measurespresumed to reflect working memory
(e.g.,averagecorrelationsof .32with a measureof the number
of digrtsthat can be correctly recalledin the reversesequence
of presentation,and .56 with a measureof the number of
digits that can be correctly recalledafter first subtractingthe
number 2 from eachdigit).
Selection of Cognitive Tasks
Severalcriteria were also consideredin the selectionofthe
cognitive tasks used to examine the predictions from the
hypothesis.One criterion was that the
tasksshould be similar to psychometricteststhat have been
consideredto representdistinct domains of cognilive activity.
The rationale for this criterion is to ensurethat the hypothesized working-memory processingresourceis truly general.
not specificto a particular type of cognitive task. A second
criterion wasthat the tasksshould allow within+ask variation
of hypothesizedresourcedemandsby quantitative.not simply
qualitative, manipulation of complexity. The requirement
that complexity be varied quantitatively was introduced to
minimize the possibility that complexity-relateddifferences
in performancecould be attributed to new processingcom-
ponentsthat were added with qualitative changesin the task,
rather than to an increasein the hypothesizedresourcedemands. And finally, the third criterion considered in the
selectionof the cognitive tasks was that the tasks should be
analyzable in order to identify the mechanisms by which
processingresourcesinfluence cognitive performance. That
is, if the processingresourceto be investigatedis workingmemory capacity,then it should be possibleto indicate how
a smaller working-memory capacity might result in lower
levelsof perfornance in thesetasks.Two taskssatisfyingthese
criteria, and employed in the presentproject, are integrative
verbal reasoningand spatial paper folding.
Sampletrials in the four conditions of the integrativeverbalreasoningtask are illustrated in Figure l. Note that one to
four premises describing a relation between two variables
were presented,followed by a question concerningthe status
ofone variable,given a specifiedchangein another variable.
Each premise and question was displayed successivelyafter
removal of the previous premise.Variableswithin the premiseswere selectedrandomly from the alphabet,and the premises,which alwaysdescribeda relation betweenadjacentletters
in the alphabet,were presentedin a random order to maximize demandson memory.
Averagedecision accuracywas expectedto decreaseas the
number of premisesin the problem increasedbecauseof the
greater demands on the limited working-memory capacity.
That is, it was assumedthat successivelylarger requirements
on working memory are imposedwith eachadditional premise becausemore earlier premisesmust be maintained during
the registrationand encodingof later premises.
The task was also designedto allow a direct assessment
the role of working-memory factorsin the complexity-related
performance decrements.This was accomplishedby examining the elfect on decision accuracyof the number of premisespresentedwhen only one of those premiseswas actually
relevant to the decision. These "one-relevant trials" are all
similar in that the sametype of decisionis required,involving
a question about the statusofone variable when the relation
between that variable and the causal variable had been de-
1 Premise
F andG do the SAME.
what will happento G?
2 Premises
M andN do the OPPOSITE.
L andM do the SAME.
what will happento N?
3 Premises
R andS clcthe OPPOSITE.
T andU do the SAME.
S andT do the OPPOSITE.
what will happento U?
4 Premises
D andE do the SAME.
B andC do the OPPOSITE.
C andD do the SAME.
E and F do the OPPOSITE.
what will happento F?
Figure l. Illustration of the sequenceof displaysfor the four conditions in the integrative-reasoning
scribedin a singlepremise,and only the context in which the
relevantinformation is presentedchanges.For example,if the
question in the four-premisecondition illustrated in Figure I
wasabout B and C, then only one premise(the second)would
be relevantto the decision,despitethe fact that a total offour
premiseshad actually been presented.
The decisionprocesses
can be assumedto remain constant
across conditions involving different numbers of premises
when only one of those premisesis relevant to the decision.
Therefore,a reduction in decision accuracywhen additional
premisesare presentedcan presumablybe attributed to a loss
of necessaryinformation from sometype of working memory.
To the extent that the slope ofthe function relating decision
accuracy to number of presentedpremisesin one-relevant
trials is similar to that basedon the data from all trials, one
could infer that reducedaccuracywith additional premisesis
largelycausedby a failure to preserveearly information during
the presentationand processingof later information.
Sample trials in the four conditions of the spatial paperfolding task are illustrated in Figure 2. Successive
this task representeda squarepieceofpaper folded from one
to four times, the punching of a hole in the folded paper,and
a pattern of circles indicating the locations of the punched
holesin the unfolded paper.Although Figure 2 illustratesonly
the outcomeof eachfold, implementation of the experimental
task on a microcomputer allowed dynamic displays of the
folds as they were in progress,rather than simply portraying
static representationsof the outcome of each fold. The task
for the subject was to decide whether the pattern of holes in
the hnal display was consistentwith the pattern that would
result from the earlier sequenceoffolds and punch location.
As in the reasoningtask, decisionaccuracywas expectedto
decreaseas the number of relevantpiecesof information was
increased.The relevantinformation in this task was presumably the type (e.g.,horizontal, vertical,diagonal)and location
(e.9.,top, middle, bottom, left, center,right) of eachfold, and
progressivelymore of this information had to be represented
as the number of folds in the problem increased.
Becausethe paper-fioldingand integrative-reasoningtasks
were designedto be structurally equivalent,direct analysesof
the role of working memory in the paper-folding task were
possiblein a manner analogousto that in the reasoningtask.
That is, memory factorswould be implicated if similar effects
of the number of presentedfolds on decision accuracywere
evident acrossall trials and on trials when only one fold was
relevant to the decision. For example, if a trial in the fourfold condition consisted of nonoverlapping folds of each
corner, as portrayed in the first two folds in the exampleof a
four-fold trial in Figure 2, then only one of the folds would
be relevant to the decision, regardlessof the location of the
punched hole. A discovery of parallel functions relating the
number of presentedfolds to decision accuracyin data from
all trials and in data from only one-relevant trials would
thereforelead to the inferencethat much of the complexityrelated performance reduction was attributable to a loss of
the relevantinformation.
Review of Predictions
We will now briefly review the predictions derived from
the hypothesisthat at leastsomeofthe age-relateddiflerences
2 Folds t]
3 Fords
L] i
F-I F_"l
u --
4 F of l' dAsV ' r i .
Figure2. Illustrationof the sequence
of displaysfor the four con_
ditionsin thepaper-folding
in certain cognitive tasks are attributable to age-relateddeclines in a working-memory processingresource.Fint, it is
predicted that the effects of the number of premisesor the
number of folds should be greaterwith increasedagebecause
older adults are assumedto havea smaller supply of available
resourcesthan young adults to cope with greaterdemandson
those resources.The secondprediction is that the magnitude
of the complexity effects in the integrative-reasoningand
paper-foldingtasksshould be correlatedwith one another if a
common processingresourcecontributesto both effects.And
finally, it is predictedthat statisticalcontrol of the computational span working-memory variable should significantly
attenuate the effects of age on the measuresof accuracy in
the reasoningand paper-folding tasks becauseage effectsin
the latter variables are presumed to be mediated by ageeffects
in the resourceconstruct indexed by the former variable.
In addition to examining these resourcepredictions, the
current project was alsodesignedto allow a decompositionof
the reduction in performanceassociatedwith an increasein
the number of premisesor folds. That is, if the decline in
accuracy associatedwith an increasein task complexity is
primarily attributableto the lossof relevantearly information
during the processingof later information, then comparable
performancereductionsshould be evident acrossall trials and
in trials when only one fold or premise is relevant to the
decision. On the other hand, if processesother than the
simultaneous retention and processingof information are
involved in the complexity-relatedperformancedeclines,then
the effectsof number of premisesor number of folds should
be much smaller in trials when only a singlepremise or fold
is relevantto the decision.
A total of 120 males,20 from each decadefrom the 20s through
the 70s, participated in the project. All but a few, who were still
students,were alumni of Georgia Institute of Technology,an insti_
tution with a primarily engineeringand technically oriented curricu_
lum. The samplecan thereforebe consideredrelativelyhomogeneous
with respectto intellectuallevel,socioeconomicclass.and educational
expenences,although there was moderatevariation in current occupation. The mean years ofeducation reported was 17.1, and this
variable was not correlatedwith chronologicalage (i.e., r: -.01).
Self-reportedhealth on a five-point scale(l : excellent,5 : poor)
averaged1.4, and 98% of the individuals reported themselvesto be
in at least averagehealth. The self-reportedhealth rating correlated
.26 (p <.01) with chronologicalage,indicating slightly poorer eval_
uation ofone's healthwith increasedage.This variablewasintroduced
as a covariatein all of the data analysesreported below. but it did
not alter any of the results and accounted for lessthan lVo of the
variancein all ofthe analyses;thus it is not discussedfurther.
The original experimentaldesignwas for eachsubjectto perform
the fiollowingtasks:the computational span task, four blocks of 40
trials each on the integrative-reasoning
task, four blocks of 40 trials
each on the paper-foldingtask, and then a repetition ofthe compu_
tational span task. However, in order to ensureat leasta minimum
amount of data from each subject in each task, a schedule was
followed in which the subject was switched to the next task after a
specifiedtime had elapsedfrom the beginning of the session.This
proved propitious becausea moderatepercentageofthe subjectswere
unableto completethe desirednumber of trial blockswithin the time
limits of 2.5 hr imposedon the session.Resultsare thereforereported
from the first two blocks of the reasoningand paper-foldingtasks
completedby everyone,and the data from subjectscompleting four
blocksare usedto examinethe reliability of the cognitiveperformance
measures.(It should be noted, however, that the same pattern of
results as that reported was also evident when the analyseswere
conductedon all of the data, with the number of trial blocks completedin eachtask as a covariate.)
Figure 3A illustratds the implementation of the computational
span task in the current project. Each frial consistedof a seriesof
presentedarithmetic problems,with the seconddigit in
eachproblem highlightedas a to-be-remembereddigit. Subjectswere
instructedto answereacharithmetic problem as it wasdisplayedand
then to recall the sequenceof target digits when the word recall
appeared.In the four-problem example illustrated in Figure 3A, the
subject should answer 5, 3, 4, and 6 to the arithmetic problems,
followed by 2-4-2-I in responseto the recall probe.Targetdigits were
randomly selectedfrom the set l-9, with the constraintsthat the
target digit was not identical to the correct answeron that problem
and that successive
to-be-remembereddigits were not identical.
Figure38 illustratesthat a double random staircasepsychophysical
procedurewas used to identify each subject'scomputational span.
That is, two independentsequences
oftrials werepresentedin random
alternationwithin the sameblock of trials. The number of arithmetic
problems in each s€quencewas increasedif the target items were
recalledcorrectly, and it was decreasedif they were recalled incor_
rectly. No changein the number of arithmetic problems fior a given
sequencewas introduced if an error was made in any of the compo_
nent arithmetic problemsfor that trial. The task wasterminatedwhen
the two sequenceshad problem lengthswithin two of eachother for
six consecutivetrials or when a total of 25 trials had been presentd.
The spanwasthe averageofthe valuesfrom the two sequences
at the
point of termination.
The data in Figure 38 are from a representativeadministration of
the task. Note that the task started with a nine-problem trial in
SequenceA, but not all nine digits were correctly recalled;thus the
next trial in that sequencehad only eight problems.The recallattempt
wasalso unsuccessfulon this trial, and so the next trial in the sequence
l 3- - _+ E =- tt ll
l 6 - l 2 l =? l
words increase,on the lower left of the screen,and decrease,on the
lower right of the screen;decisionswere communicated by pressing
the *Z' or " /" key on the keyboard,respectively.In the paper-folding
task, the display of the pattern of holes w:rs accompanied by the
words zo on the lower left ofthe screenand yes on the lower right of
the screen;decisionswere communicatedby pressingthe "Z" or " /"
key, respectively.
l 5* 1 1 =
f REcALL--]
t - - - - l
F 8
8 4
2 3
6 I 10 12 1416
Figure 3. Panel A: Illustration of the sequenceof displays in the
computational span task. Panel B: Example of how the span is
estimated from the convergenceof two independent sequencesof
contained only seven problems. However, the seven-problemtrial
wasnot pres€nteduntil Trial 4 becausetherewasa switch to Sequence
B for Trial 3. This sequencestartedwith a two-problem trial, but an
arithmetic error was committed on that trial; thus the next trial in
this sequence,Trial 5, also contained only two arithmetic problems.
Eventually the two sequencesconvergedto the specified criterion,
yielding an estimatedcomputational span of six items.
All phasesof the task were self-pacedin that the next arithmetic
problem within a trial was not presenteduntil an answer had been
enteredon the computer keyboardto the previous problem, and the
next trial was not presenteduntil the appropriate number of digits
had been enteredduring the recall phasefor that trial.
The reasoningand paper-foldingtaskswere presentedin blocks of
40 trials. Each block contained l0 trial typ.:s consistingof n : 1,2,
3, or 4 premisesor folds, with from I to n premisesor folds relevant
to the decision.In other words, there were four trial typescontaining
four premisesor folds (with l, 2, 3, or 4 relevantpremisesor folds),
three trial types containing three premisesor fiolds(with l, 2, or 3
relevant premisesor folds), and so forth. Each block contained a
random arrangement of two positive (increase or yes) and two
negative(decreaseor no) trials of eachtype. Feedbackindicating the
correct answer in the trial was displayed after each response.
A trial was initiated in both tasksby pressingthe enter key on the
computer keyboard.The first fold or premisewasthen displayed,and
each successivepremise or fold was displayed by pressingenter aga:in.
The question display in the reasoningtask was accompaniedby the
GeneralAge Elfects
Initial analysesfocusedon the variablesof mean computational span (i.e., the averageof the two spansfor a given
individual, designatedCSpan) and mean percentagecorrect
acrossall 80 trials in the fint two trial blocks for both the
reasoningand paper-foldingtasks.Averagesofthese variables
acrossthe 20 subjectsin each age decadeare displayed in
Figure 4. The most striking feature of this figure is that the
age trends are remarkably similar acrossthe three variables.
In fact, the slopesof the linear regressionequations relating
ageto p€rformancewere -.43%/year for both reasoningand
paper folding, and -.04 digits/year for CSpan.
Table I contains the matrix of correlations among the
primary variables,with estimatedreliabilities in parentheses
along the diagonals. Reliability of the CSpan measurewas
obtainedby boostingthe correlation betweenthe valuesfrom
the first and second computation span assessments
by the
Spearman-Brownformula to estimate the reliability of the
averagevalue. The reasoningand paper-folding reliabilities
are correlations between the values from the first two trial
blocks and the valuesfrom the last two trial blocks for the 49
subjectswith four blocks in the reasoningtask and for the 55
subjectswith four blocks in the paper-foldingtask.
The data summarized in Figure I and Table I were subjected to hierarchical regressionanalysesto examine the effectsofage on reasoningand paper-foldingperformanceafter
partialing out the variance associatedwith the CSpan measure. For the reasoningtask, the R2 attributable to agewhen
it wasconsidered
alonewas .278,F(1, I 18): 45.51,MS":
1 3 6 . 4 8p, < . 0 1 , a n d t h i s w a s r e d u c e dt o . 1 1 9 ,F ( 1 , 1 1 7 ) :
21.41,MS.: 123.95,p < .01, after removingthe variance
associatedwith the CSpan working-memory index. Similar
-r- Reasoning
-r- PaperFolding
E 70
d 6 0
6 F
5 { )
45 55
Figure 4. Mean values of computational span task (CSpan) and
averagepercentagecorrect in the reasoning and paper-folding tasks
for the 20 individuals in eachagedecade.
Table 1
Correlations Among Primary Variablesfor all 120 Subjects
acrossall trials and acrosstrials when only one premise was
relevantto the decision.Meansacrossall 120subjectsof these
valuesare illustrated in Figure 5. As expected,increasingthe
number of premisespresented,and thus presumablyincreas_
ing the demands on the limited working_memorycapacity,
2. CSpan
resultedin progressivelylower levelsofaverage accuracy.Of
4. Paperfolding
particular interest,however,is the similarity in ttre funciions
for all trials (i.e., a slopeof -g.g% per premise)and for trials
when only one premise was relevant to the decision (i.e.. a
slope of -7.4% per premise).On the basis of the argument
Note. All correlationsare significantat p < .01.
outlined earlier,thesefindings can be interpretedassuggesting
CSpan: computational span.Estimatedreliabilitiesare in parentheses.
that much of the performance decline associatedwith in_
creasedcomplexity in this task is attributable to a failure to
pres€rverelevant information in the more complex
resultswere evident in the paper-foldingtask becausethe R2
for age,when consideredalone, was.2g2,F(1, I lg) = 46.45,
Analysesof the effectsof age on these complexity slopes
MS,: 136.59,p < .01, and this was reducedto .tOt, f1t,
were restrictedto subjectswith an accuracyof at leastgSV"in
l17) 27.99,MS.: 133.20,p < .01, after statisticalcontroi
the one-premisecondition in order to minimize the possibility
of the CSpan measureof working-memory capacity.
of floor effectsin the slopemeasures.That is, low performancl
Quantitative estimatesof the relative importance of the
in the simplest condition precludesmeaningful estimatesof
CSpan-indexedworking-memory factor to the agedifferences
the complexity slopesbecauseof insufficient rangeofperform_
in the verbal integrative-reasoningand spatial paper-folding
ance variation from the one-premise to the four_premise
taskscan be derived from the resultsjust described.That is,
condition. The value of 85Vowas somewhatarbitrary; it was
the proportional reduction in age-associatedvariance
high enough to allow an adequaterange for lower accuracy
achieved by statistical control of the CSpan index can be
when more premiseswere presented,but yet not too high t;
determined by dividing the differencein age-associated
exclude a large proportion of subjects.The correlation be_
ance with and without the control of CSpan by the total
tween age and accuracy in the one_premisecondition was
amount of age-relatedvariance.Thesecomputationsrevealed
-.30, and thus more old than young
adults were excluded
that statisticalcontrol of the CSpanindex of working_memory
with this procedure.The restrictedsample of 104 subjectsis
capacity attenuatedthe age e{fectsin the reasoningtast by
still representativeof the larger sample,Lo*eue., becausethe
57.27o,and it attenuatedthe age effectsin the paper-foldin!
correlation betweenage and averagepercentagecorrect was
taskby 42.9%.Although the absoluteproportionsdiffer acrosi
--48 in this group compared with -.53
in the total sample.
the two tasks,the results are consistentin suggestingthat a
The mean agein the restrictedsamplewas 46.9 years(SD =
moderateproportion ofthe ageeffectsin both reasoningand
16.4)comparedwith 49.3 years(,SD: 16.g)for the entire
paper-folding performance is mediated through age_related
reductions in working-memory capacity, as the latter is in_
Reliabilities of the complexity slope parametersrelating
dexedby the CSpan variable.
decision accuracy to number of presented premises were
estimatedby determining the correlation betweenthe slopes
Practice E/fects
from the frrst two trial blocks and from the last two trial
blocks for subjectswith accuracyofat least g5% in the one_
In all three tasks,practice resulted in better performance,
premisecondition in both setsof trials. The correlationsfor
but it did not substantially alter the magnitude of the age
the 4l subjectswith the appropriate data were .76 for the
relations. That is, the first computational span assessment
yreldedan averagevalue of 5.65, with an age correlation of
-.39, while the secondassessment
resulted in an averageof
6.21, with an age correlation of -.45. practice effectsin the
r- All
reasoning and paper-folding tasks were examined for the
subjectscompleting four trial blocks in each task. The 49
subjectswith four blocksin the reasoningtask averagedg0.3%
o B o
correct decisions for the lirst two blqcks (age correlation :
-.58) and averaged82.8Vofor
the last two blocks (agecorre_
7 0
lation : -.46). The 55 subjectswith four blocks in it " pup"r_
folding.task averaged 78.1% in the fint two blocki (age
correlation : -.44) and B2.0Voin the secondtwo blocks
I nt egtat iv e- Reasoning pedormanc e
Decision accuracy in the integrative-reasoningtask was
computed as a function of the number of premisespresented
Numberof premises
Figure 5. Mean percentage correct as a function of number of
premisespresented for all trials and for trials with only one relevant
premisein the reasoningtask.
slope basedon all trials, and .56 for the slope basedon the
Mean slopesfor the 13-20 subjectsin each agedecadeare
illustrated in Figure 6. Despitethe apparentfloor effectbegin_
ning in the decadeof the 50s,the agecorrelationswere -.46
for the slope basedon all trials and -.42 for the slope based
on only one-relevanttrials. Thesevery similar pattern55rrggest
that the composition of the complexity effect is relatively
constant acrossthe adult years.Stated somewhatdifferently,
the fact that nearly parallel age trends were evident in the
complexity slopesfrom all trials and from only one-relevant
trials can be interpreted as indicating that most of the agerelated increasein the effectsof additional premisesis attrib-.
utable to age-relatedincreasesin the loss of relevant information, that is, to working-memory limitations.
Results from the hierarchical regressionanalyseswith the
slope basedon all trials were that the R2 attributable to age
when it was consideredalone was.212, F(1, 102\ = 27.3j,
MS": 25.31,p < .01, and the increment in R2 associated
with ageafter controllingfor CSpanwas.148,f'(1, l0l):
1.9.20,MS, = 25.23,p < .01. Correspondingresultsfor the
slopesbasedon one-relevanttrials were that the R2 attributableto ageby itselfwas.174, F(|, 102): 21.46,MS. : 38.I I ,
p < .01, and the increment in R2 associatedwith age after
controllingfor CSpanwas .141,f'(1, 101)= 17.36,MS.:
38.47, p < .01. Expressedas measuresof percentageof attenuation due to statisticalcontrol ofthe CSpanindex, the values
were 30.2% for the slope from all trials and, l9.0%ofor the
slope from one-rel€vanttrials.
Paper- Folding Performanc e
Analysessimilar to those conducted on the measuresof
reasoningperformance were conducted on the measuresof
paper-foldingperformance.Mean decisionaccuracyasa function of the number of folds presentedis displayedin Figure 7
for all trials and for trials with only one relevantfold. Regression slopesbasedon the averagedata were -6.9Vo per fold
with all trials and -5.3% per fold with one-relevanttrials.
Figure 7 revealsthat accuracywith one-relevantfold in the
three-fold condition app€ars abnormally high, and it was
suspectedthat this might be due to the existenceof a hieh
'E -o
- . g - O
Y ( D
dE -g
o -i2
45 55
Figure 6. Slopes of the functions relating percentagecorrecr ro
number of premisespresentedfor all trials and for trials with only
one relevantpremisefor the I 3-20 individuals in eachagedecade.
r- All
,'' 1-Relevant
6 9 0
8 e o
s 7 0
b 6 0
Numberof Folds
Figure7. Meanpercentage
correctasa functionof numberof folds
for all trialsandfor trialswith onlyonerelevantfold in the
proportion of either simple folds or obviousdistractorsin this
condition, resulting in a set of particularly easy problems.
However, ratings from four judges of the perceiveddifficulty
of the one-relevantproblems increasedmonotonically wittr
additional folds and did not correspond to the observed
performancepattern.We arethereforeunableto offer a reason
for the nonmonotonic trend in the one-relevantdata.
Age effects in the complexity slopeswere examined after
first excluding subjectswith accuraciesof less than g5% in
the one-fold condition. The rationale for this restriction is the
same as that in the analysesof the reasoning data. The
correlation of -.32 betweenageand percentagecorrect in the
one-fold condition indicatesthat a larger number of old than
young adults were eliminated with this procedure, but the
restrictedsampleis still representativeof the completesample
becausethe correlation betweenage and averagepercenuge
correct was -.53 in both samples.The mean age of the 7g
subjectsin the restrictedsamplewas 45.3 (SD = 15.3),comparedwith the mean of 49.3(SD = 16.8)for all 120subjects.
Reliabilities of the slope parameterswere estimated from
the correlations between the values from the hrst two trial
blocks and the last two trial blocks for the 36 subjectswith
averageaccuraciesin the one-fold condition of at least g5%
in both setsof trials. The correlationswere .42 for the slopes
from all trials and .39 for the slopesfrom one-relevanttrials.
Mean slopesfor the 7-17 subjectsin each age decadeare
displayed in Figure 8. The similar age trends for the two
slopes,togetherwith the nearly identical agecorrelations(i.e..
r: -.47 for slopesfrom all trials, r : -.45 for slopesfrom
one-relevanttrials), suggestthat much of the age-relatedincreasein the elfectsof task complexity is attributable to a loss
of relevant information from working memory.
Hierarchicalregressionanalysesrevealedthat the R2 in the
complexity slope from all trials attributable to agewhen age
w a s c o n s i d e r eadl o n ew a s . 2 1 8 ,F ( 1 , 7 6 ) : 2 1 . 1 5 , M S . :
15.88,p < .01, and the incrementin R2 associated
with age
afterthe control of CSpanwas.177, F(|, 75) : 17.34,MS. :
15.77, p < .0 I . Correspondingvaluesfor the slope from onerelevanttrials were .199 for R2 due to agealone,F(1,76) :
18.89,M,S": 41.76,p < .0l,and .158,F(t, 75): 15.t9,MS"
: 41.31,p < .01, for the increment in R2 due to age after
control of CSpan. The percentageof attenuation of the age
effects by control of the CSpan index of working-memory
doO *
-+ All
'''t' 1-Relevant
45 55
Figure 8. Slopesof the functions relating percentagccorrect to
numberof folds presentedfor all trials and for trials with only one
relevantfold for the 7-17 individualsin eachagedecade.
capacitywas therefore 18.8%for the slope from all trials and
20.6Vofor the slope from one-relevanttrials.
of Performance Measures
Correlationsamong all relevant measuresfor the 73 individuals with accuracyof at least 85% in both the one-premise
reasoningcondition and the one-fold paper folding condition
are presentedin Table 2. Of particular interest in this table
are the correlations of .63 between the complexity slopes
based on dl trials and .45 between the complexity slopes
based on one-relevant trials. These values are impressive
becausein both casesthey are close to the averageof the
estimated reliabilities of each measure(i.e., .76 and .42 for
the slopesfrom all trials, and .56 and .39 for the slopesfrom
one-relevanttrials). It thereforeappearsthat nearly all ofthe
reliable variance in the complexity slope measuresfrom the
two tasksis shared,or common, variance.
Three predictions, derived from the view that one determinant of age differencesin certain reasoning and spatial
ability cognitive tasksis an age-relatedreduction in a workingm€mory processingresource,were examined in this study.
The results provide strong support for two of the predictions
and moderatesupport for the third prediction.
One prediction was that if cognitive performancedeclined
with increasedtask complexity becauseof greater demands
on a limited processingresource,then the magnitudeof those
performance declines should increase with increased age becauseofthe hypothesizedage-relatedreduction in processing
resources.The resultssummarizedin Figures6 and 8 indicate
that this prediction was clearly confirmed. Furthermore, the
discovery that very similar age trends were evident in the
complexity slopesderived from trials in which only a single
premiseor fold wasrelevantto the decisionsuggests
that most
of these age effects can be attributed to a failure to retain
information from early premisesor folds during the presentation of subsequentpremisesor folds. In this respect,limitations of the working-memory processingresourceappearto
be implicated as a major determinant of the age-complexity
effectsin the presentstudy.
A secondprediction examined in the study was that if the
complexity-relatedperformancedeclineswere attributable to
limited quantities of a relevant processingresource,then the
magnitude of those declinesfor a given individual should be
similar acrossdi{ferent tasks involving that same resource.
The complexity manipulations in the current reasoningand
paper-foldingtaskswereassumedto involve a resourcerelated
to working-memory capacity, and thus it was expectedthat
the slopesof the functions relating accuracyto complexity in
eachtask would be significantly correlatedwith one another.
This expectation was confirmed in that the correlation between the two slopes based on all trials was .63, and the
correlation between the slopesfrom one-relevanttrials was
.45. Becausethe correlationwith the other measurewasnearly
as high as the expectedcorrelation of the measurewith itself
(i.e., the averageof the estimatedreliabilities was .59 for the
slopesfrom all trials and .46 for the slopesfrom one-relevant
trials), it can be concludedthat the two measuressharenearly
all oftheir reliable variance and, hence,probably reflect the
Table 2
Correlations Among Select Variablesfor 73 Subjects With at Least 85% Conect in the One-PremiseReasoning Condition and
in the One-Fold Paper Folding Condition
- 8 . 15
Nole. CSpan is computational span; R%C is averagepercentagecorrect in the reasoningtask; PFToCis averagep€rcentagecorrect in the pap€rfolding task; RS and RIS are complexity slopesfor all and one-relevanttrials in the reasoningtask; and PFS and PFIS are complexityilopes
for all trials and for one-relevant trials in the paper-folding task.
Correlations greater than .30 are significant at p < .01. Estimated reliabilities are in parentheses.
Althoug;h the results relevant to the first two predictions
suggestthat working-memory factors are involved in the
tendency for age differencesto increaseas task complexity
increases,signihcant age effectswere also found in the least
complex (i.e., one premise and one fold) conditions. A plausible inference from this combination of findings is that at
leasttwo factorscontribute to the agedifferencesobservedin
the current integrative-reasoning
and paper-foldingtasks:limitations of working memory and some as-yet-unspecified
factors that are responsiblefor the differencesobservedwhen the
tasksappearto place little or no demandson working mem-
Resultsfrom the analysesof the prediction that statistical
control ofan index ofprocessing resourceswill attenuatethe
age differencesin measuresof cognitive performance also
seem consistent with this dual-determinant interpretation.
That is, although partialing out the variance associatedwith
the CSpan measureof working memory attenuatedthe age
effectsin the measuresof overall percentagecorrect in the
reasoningand paper-foldingtasks,it did not completelyeliminate them.
These results are subject to at least two quite distinct
interpretations.On the one hand, it could be arguedthat the
discovery that at least some of the age differences in the
presentreasoningand spatial tasksappearto be mediatedby
age-relatedreductions in working memory, as the latter is
indexed by the CSpan measure,representsan important step
toward understandingwhy age differencesoccur in certain
cognitivetasks.According to this perspective,there are almost
certainly multiple determinants contributing to the age-related differencesin cognitive functioning, and consequently
it would have been unrealistic to have expected complete
elimination of the agedifferencesby statisticalcontrol of any
singlevariable.The fact that an appreciableproportion ofthe
agedifferencescan apparentlybe attributed to a factor related
to working memory might thereforebe viewed as an impressive outcome.
On the other hand, it could also be arguedthat the failure
to eliminate, or at leastdramatically attenuate,the agedifferencesafter statistical control of the index of the workingmemory processingresourceis inconsistentwith the presumed
importance of the processing-resource
construct. That is, the
resultsofthis study and ofseveralrelatedonesinvolving other
combinations of cognitive measuresand resource indexes
(e.9.,Salthouse,1988a,1988b;Salthouse,Kausler,& Saults,
1988) suggestthat less than 50%, and perhaps only l5Vo20%, of the overall agedifferencesin fluid cognitivetaskscan
be accountedfor by age-relatedreductions in processingresourcesas assessed
with available measures.Becausethe assumption implicit in many discussionsis that an age-related
reduction in processingresourcesis a major determinant of
the age differencesobservedin certain cognitive tasks,these
results,which at best suggesta relatively weak influence of
processingresources,might be viewedasrather disappointing.
Although debatescould obviously continue concerningthe
validity of each of these interpretations, a more productive
focus for future efforts might consist of carefully examining
reasonswhy there was not greater attenuation of the age
effects after statistical control of the measure of working
memory capacity.Of course,one possibility is that the atten-
uation was small becauseworking-memory factorsplay only
a minor role in the agedifferencesin the current measuresof
cognitivefunctioning. This view is clearlyplausiblewith some
variables, but it seems unlikely with the measuresin the
presenttasks,particularly the slopemeasuresrepresentingthe
amount of changein accuracywith each additional premise
or fold. That is, the nearly equivalent slopesfor all trials and
for one-relevanttrials indicate that accuracydecreaseswith
additional premisesor folds in large part becauseof an inability to preserverelevant information during the processing
of other information, a situation that appearsto epitomize
failure of working memory.
A secondpossiblereasonfor the relativelysmall attenuation
of the ageeffectsby statisticalcontrol of the working-memory
resourceis that cognitive performancemay not be a simple
linear function of the quantity or capacity of working memory. For example, there may be something analogousto a
resourcethreshold wherebycognitive performanceis not impaired if the working-memory resourceexceedssome minimum value, but it is severly impaired as the resourcefalls
below the minimum or threshold value. Although the existence of a threshold would normally be expectedto yield a
distinctive stepfunction, patternsof this type may be diflicult
to identify in group data if people vary in the positions of
their respectivethresholds. Therefore, unless it can be assumed that the relation betweenresourcequantity and cognitive performanceis reasonablylinear and relatively similar
across age groups, linear regressionprocedures may yield
misleadingestimatesof the resourcecontribution to the agerelatedcognitive differences.
Another factor that may have contributed to the relatively
modest attenuation of the age effectsafter statisticalcontrol
of an index of the working-memory processingresourceis the
possibility that working memory was not adequatelyassessed
by the CSpan measure.At least three issuescan be raised
concerningthe validity of the CSpan measureas an index of
working memory. One is relatedto the fact that even though
severalcriteria were consideredin selectingthe CSpan measure, it still has all of the limitations associatedwith singlevariableassessment
of a complex construct (e.g.,narrow and
test-specificreflection of construct,restrictedreliability, etc.).
Had suitable alternative measuresof working memory been
available, these limitations could have been minimized by
using multiple indicators of the working-memory construct.
A secondpotential weaknessof the CSpan measurewith
respectto its validity asan index of working memory concerns
the possibility that the measuremay not have reflectedthe
samecharacteristicin every subject.The CSpanmeasurewas
selectedover alternative measuresbecauseit allows accurate
monitoring of both processing(accuracyof arithmetic computations) and storage(accuracy of digit recall), and it has
respectablereliability when assessed
with appropriate procedures. However, observation of the subjectsperforming the
computation span task revealedthat at least some subjects
attemptedto rehearsethe to-be-remembereddigits during the
time in which arithmetic problemswereto be solvedasrapidly
as possible.In fact, analysesof the time for each arithmetic
problem during the computation span task indicated that an
averageof approximately 4 s was spent on each problem,
compared with the less than 2 s needed by pilot subjects
performing the arithmetic problems without any concurrent
memory requirement.Furthermore,the fact that considerably
smaller spans are obtained when the task is administered
under experimenter-pacedconditions ofabout 2 s per problem (Salthouse,1988b;Salthouse& Prill, 1987)also suggests
that deliberate rehearsal strategiesmay have been used by at
leastsome of the presentsubjects.
The possibility that subjectsmight have varied in the manner in which they performed the task clearly complicates
interpretation of the resulting measuresas reflections of a
working-memoryprocessingresource.The nature of this complication can be elucidatedby thinking of working memory
as involving the two simultaneoustasks of storageand processing.As with other dual-tasksituations,therefore,comparisons of performance in the measuresof one task are most
meaningful only if there is little or no variation in the performanceof the other task. Becausethe lengthyaveragetimes
taken to perform the arithmetic (processing)task were also
associatedwith considerablebetween-subjectvariability, it is
likely that subjects differed in their relative emphaseson
storagevenus processing,and thus the CSpan measuremay
not have reflectedthe same property of working memory in
every individual. One means of investigatingthis interpretation, currently under way in our laboratory, involves measuring the efficiency ofprocessing and the capacity ofstorage
in both single- and dual-task situations and then examining
the relative emphaseson processingas opposedto storagein
the dual-task,or working-memory, situation.
Still another issue concerning the validity of the CSpan
measurerelatesto the manner in which working memory is
most appropriately assessed.
That is, it could be arguedthat
working memory implies that the memory is actually used
during the perfiormanceof a relativelycomplex cognitivetask.
From this penpective,therefore,the bestmeasuresof working
memory may not be those derived from a task specifically
designedto measurea particular ty,peof memory capacity,
within the context
but rather thoseobtainedfrom assessments
of on-goingcognitive activities.
In the presentstudy, the slope measuresrepres€ntingthe
changein accuracyfor one-relevanttrials acrossincreasesin
the number of presentedpremisesor folds appearpromising
as potential candidatesfor this type of within-context working-memory assessment.
The discoverythat the valuesfrom
the verbal reasoning task were significantly correlated with
those from the spatial paper-folding task is also encouraging
becauseit suggeststhat the measuresreflect a general, rather
than task-specifrc,construct. Unfortunately, the assessment
proceduresused in this study resultedin the slope measures
having relativelylow reliability, and thereforetheir usefulness
in statistical-controlanalysesis somewhatlimited. Reliability
could undoubtedly be improved by increasingthe number of
appropriate observations from each subject, however, and
further investigation of within-context working-memory assessmentdefinitely appearswarranted.
In conclusion,the resultsof the presentexperimentappear
reasonably consistent with the processing-resourceinterpre-
tation of age differences in cognitive functioning. Although
this perspective is still uncomfortably vague, it seems to
provide a parsimonious account of three results that are
currently diffrcult to explain without the processing-resources
construct: (a) increases with age in the magnitude of complexity-related performance differences; (b) high correlations
across subjects in the magnitudes of the complexity slopes
from the verbal integrative-reasoning task and the spatial
paper-folding task; and (c) moderate attenuation of the age
effects with statistical control of an index of the workingmemory processing resource. It is clear, however, that greater
understanding is needed of the nature of working memory,
and how it can best be measured. before a consensus can be
expected with respect to the contributions that age-related
reductions in working-memory processing resources might
make to age differences in cognitive performance.
Baddeley,A. (1986). Working memory.New York: Oxford University
Baddeley,A., Logie,R., & Nimmo-Smith,I. (1985).Componentsof
fluent reading.Journal ofMemory and Language,24, I l9-131.
Case, R. (1985). Intellectual developmentfrom birth to adulthood.
Orlando, FL: Academic Press.
Case,R., Kurland, M., & Goldberg,J. (1982).Operationalefliciency
and the growth of short-termmemory span.Journal of Experimental Child Psychology,33,386-404.
Cattell, R. B. (1971). Abilities: Their structure,growth, and action.
New York: Houghton Mifllin.
Craik, F. I. M., & Rabinowitz, J. C. (1984). Age differencesin the
acquisitionand useof verbal information: A tutorial review. In H.
Buouma & D. C. Bouwhuis (Eds.),Attention and performanceX
(pp. 47 | 499). Hillsdale, NJ: Erlbaum.
Daneman, M., & Carpenter,P. A. (1980). Individual differencesin
working memory afld reading. Journal of Verbal Learning and
VerbalBehavior.I 9. 450-466.
Hebb, D. O. (1942). The effect of early and late brain injury upon
testscores,and the nature ofnormal adult intelligence.Proceedings
of the American PhilosophicalSociety, 85, 275-292.
Horn, J. L. (1982). The theory of fluid and crystallizedintelligence
in relation to conceptsofcognitive psychologyand aging in adulthood. In F. I. M. Craik & S. Trehub (Eds.),Aging and cognilive
processes(pp. 237-278). New York: Plenum.
Salthouse,T. A. (1985). A theory of cognitive a8'in8'.Amsterdam:
Salthouse,T. A. ( I 9E8a).Resource-reductioninterpretationsof cognitive aging. Developmental Review, 8, 238-27 2.
Salthouse,T. A. (1988b).The role ofprocessingresourcesin cognitive
aging. In M. L. Howe & C. J. Brainerd (Eds.),Cognitivedeveloy
ment in adulthood(pp. 185-239).New York: Springer-Verlag.
Salthouse,T. A., Kausler, D. H., & Saults,J. S. (1988).Utilization of
path-analytic procedures to investigate the role of processing resourcesin cognitive aging.Psychologyand Aging, J, 158-166.
Salthouse,T. S., & Prill, K. (1987).Inferencesabout ageimpairments
in inferential reasoning.Psychologyand Aging, 2,43-51.
Welford, A. (1958).Ageing and human skill. London: Oxford University Pres.
ReceivedOctober 26, 1987
Revision receivedSeptemberl, 1988
AcceptedSeptemberl, 1988 r
Related flashcards


29 cards

Psychological concepts

24 cards


38 cards

Secret societies

16 cards

Religious titles

17 cards

Create Flashcards