Copt'right 199J bt The GcftililobRieil
J ournal of Geront olo gy : PSYCHO LOG IC AL SCI ENC ES
1 9 9 3 .V o l . 4 8 . N o . 5 . P 2 4 5 P 2 5 5
Sodtn ol Aneriut
Influenceof Task-Specific
Processing
Speed
on Age Differencesin Memory
Timothy A. Salthouseand Vickv E. Coon
School of Psychology, Georgia Institute of Technology.
Two studies were conducted to investigate the aspect(s) of processing involved in the hypothesized speed mediation of
adult age differences in memory. Both studies involved a serial memory task in which information was to be recalled
either in the original order of presentation, or in a reordered sequence. Results from both studies indicated that taskspecificprocessing durations were slower among older adults thon among loung adults, but that the attenuation of the
age-related variance in memory was nearly as great after statistical control of a task-independent speed measure as
after conlrol of nsk-specffic speed measures. These findings are consistent with the hypothesis that a substantial
proportion of the adult age-related differences in memory is associated with a decrease with increased age in the speed
of executing many cognitive processes,and not simply the speedof one or two specific processes.
'JHE goal of the studiesdescribedin this report was to
r investigatehow processingspeedcontributesto the
relationsbetweenageand memory. The relationsof interest
havebeendocumented.andsomeinitial evidencerelevantto
this issuehasbeenprovided,with datafrom earlierresearch
(Salthouse,1983, in press).The memory tasksin the previous studieswere free recall of two lists of, l2 unrelated
nounspresented
auditorilyat a rateof -2 seconds/word,
and
paired associatcrecall <lf six pairs of unrelatednouns also
presentedat -2 secondsiword.The speedmeasureswere
derived from three paper-and-penciltests of perceptual
speed;digit symbol substitution,letter comparison,and
patterncomparison.The averageof the z-scoresin the three
testsservedasa compositeindex of the speedwith which the
individualcould executeelementaryoperations.
The results of these earlier studies are summarizedin
Table I . Notice that in the first study there was a large
attenuationof the age-relatedvariancein the memory measureswhen the varianceassociated
with the processingspeed
measurewas statisticallycontrolled(i.e., 87o/oattenuation
with free recall, from an R' for age of . 162-.O2l , and 857o
attenuationwith paired associates,from an R2 for age of
.162-.024).
An extreme-groupsdesignwas usedin the secondstudy,
which resultedin largerage-relatedeffectsthan thosein the
first studybecauseof the omissionof the varianceassociated
with the intermediateages. [The memory data were not
reported in the Salthouse(1993) article, but most of the
subjectsfrom the second study in that project performed
thesetasksaftercompletingthe tasksreportedin the article.l
The speedand memory taskswere identical to those of the
first study, but also includedwere measuresof associational
fluencythat might be expectedto be importantin the performanceof many memory tasksif effectivenessof elaborative
encoding is related to the number of relevant associations
that can be producedin a limited time. The measureswere
from tasks requiring the generationof as many words as
possible,beginningwith one of two stems(pro- and sub-)or
ending with one of two terminations(-ay and -ow), and the
averageof the z-scoresin the two tasksservedasthe measure
fl uency.
of associational
It is apparentin the bottom panelof Table l that the major
findingsof the first studywerereplicated,as therewas9l7o
attenuationof the age-relatedvariancein free recall performance and 88o/oattenuationof the age-relatedvariance in
paired associatesperformanceafter control of the same
compositespeedmeasure.However, it was surprisingthat
the attenuationof the age-relatedvariancein memory was
smalleraftercontrolof the presumablymore specificfluency
measurethan after control of the more generalspeedmeaattenuationwith fiee recall perforsure;that is, only 44.5o/o
performanceand 37.l%oattenuationwith pairedassociates
mance. Taken together,the resultsof theseearlier studies
suggestthat processingspeed,as indexedby variousperceptual comparisontasks, is involved in the mediationof age
differencesin memory, and that the speedinfluenceappears
to be fairly general,in that it doesnot seemto be restrictedto
measurespostulatedto reflect memory-relevantprocesses.
The existenceof significant residual age-relatedvariance
even when the variation in speedis controlledindicatesthat
other factorsalsocontributeto the agedifferencesevidentin
thesetasks.Nevertheless,the discoverythat influencesindependentof speedare associatedwith <157o of the total agerelatedvariancefor free recall and pairedassociatemeasures
of memory suggeststhat someaspectof speedplays a major
role in the relationsbetweenage and memory.
The purposeof the currentproject was to conducta more
detailed examination of the relation between speed and
memory with a simplertask in which it might be possibleto
identify the mechanismsresponsiblefor the apparentspeed
mediation of age-relateddifferencesin memory. The researchstrategyconsistedof three distinct stepsor phases.
The first step was to determine the amount of age-related
variancein the criterion measureof memory before and after
controlling the variance in an index of processingspeed
derivedfrom independenttasks.The goal of theseanalyses
P245
P246
SALTHOUSE
AND COON
Tablel. Regression
Results
FromPrevious
Studies
Free Recall
Cum.
R2
Equation
Incr.
R2
F fbr
[ n c r .R r
PairedAssociates
Cum.
R2
lncr.
R2
S a l t h o u s (ei n p r e s s ) - ( N : 3 0 5 . a g e 1 9 - 8 4 , m e a na g e : - 5 1 )
I
Age
.162 .162 58.67E
.t62
.162
2
Speed
Age
S a l t h o u s (e1 9 9 3 )3
Age
.225
.246
.225
.02'l
90.07+
lJ.53i'
.211
.238
.214
.024
F fitr'
I n c r .R l
58.65*
84.57+
9.53i.
( N = 7 1 , a g e l 9 2 6 . N - 69. age57 {19)
.546 .546 173.25+
.596 .596 212.071'
zl
Speed
Age
.51t3 .518
.567 .049
110.92+
1 6 .| 2 +
5
Fluency
Age
.297
.600
1 0 6 .l - 5 i '
l0t3.,1tiN'
6
Spced
Flucncy
Age
.51{i .5lU llJ4.43*
.-536 .01{t
6.37't'
.60 |
.06-5 23.32N'
.297
.303
.540 .540 l 9 ' 7 . 1 6 +
.609 .069 2 5 . 1 7 *
.247 .247 93.2I .r.
digit stimuli and alphabetical
order with letterstimuli). Of
particularinterestis the conditionin which reorderingof the
stimuli is required becausethe necessityof simultaneous
processingand storagesatisfiesthe criterion of a working
memory task, and working memory has beenimplicatedas
an importantfactor in the age differencesin many cognitive
tasks (Salthouse,1990). Becausethe amountof required
processingis greater with reordered sequences,the age
differencemight be expectedto be larger in that condition
relative to recall in the original order if some of the agerelateddeficit is attributableto less efficient processingon
the part of older adults(Babcock& Salthouse,1990;Gick,
C r a i k , & M o r r i s , 1 9 8 8 ;M o n i s , G i c k , & C r a i k , 1 9 8 8 ;
S a l t h o u s e1. 9 9 0 :S t i n e& W i n s f i e l d .1 9 8 7 ) .
. 6 2 2 . 3 7 5 1 4 1. 7 3 r ,
.5.{0 .5,+0 2 0 3 . 3 U i .
.541 .003
1.42
.623 .0U0 30.06i
C u n r . = c u m u l a t i v eI:n c r . : i n c r e m c n t .
, r p< . 0 5 .
was to obtain an estimateof the magnitudeof the overall
speedinfluenceon the age-memoryrelations,in a manner
analogous
to that describedabove.
The secondstep in the researchstrategyconsistedof
identifyingnreasures
of task-specific
processing
speeds,and
thenexaminingthe relationsof thesemeasures
to the criterion memorymeasure,and determiningthe extentto which
theage-related
variancein thememorymeasurewasreduced
afier controlling the variance in the task-specificspeed
measures.The rationalefor these analyseswas that the
mechanismsby which speedinfluencedrelationsbetween
age and memory might be betterunderstoodby examining
thedurationof variouscomponents
presumedto be involved
in a given memory task. That is, in order lbr processing
speedto influenceperfbrmanceit must af'tectquantitativetlr
qualitativeaspectsof processing.and informationaboutthe
durationof particularprocessingcomponentsmay indicate
which aspectsof processingare affected in what manner.
Also includedin this phasearecomparisons
of the influence
of the task-specificprocessingspeedmeasuresbefbre and
after controlling the variancein the speedmeasuresderived
from independent
tasks.Thesecomparisons
areexpectedto
be informativeaboutthe extentto which the influenceof the
task-specificspeedmeasuresis unique, and distinct fiom
that associatedwith the speedmeasuresobtainedfrom independenttasks.
The third step in the researchstrategyinvolved using the
task-specificspeed measuresas the criterion or predicted
variable,and age and the speedmeasuresfrom independent
tasksas the predictorvariables.Of interestin theseanalyses
was the amount of unique age-relatedvariancein the taskspecificprocessingspeedmeasures,as inferredby the residual age-relatedvarianceafter control of the variation in the
speedmeasuresderivedfrom independenttasks.
Two studiesare reported,both involving a serialmemory
task in which (immediate)recall is eitherin the original order
of presentation,or in a different order (numericalorder with
SI"AV I
Individualdigits or letterswere presented
at a fixed rate,
but subjectswere encouragedto respond(by typing on a
keyboard)both rapidly and accurately,and the interval
betweeneachsuccessive
keystrokein recallwas monitored
to obtainmeasures
of the durationof task-relevant
processing. Becausethe instructionabout the order in which the
itemswereto be recalledwasnotpresented
until afterthe last
item in the sequence,the time requiredto conduct the
processingassociatedwith reorderingthe items c<luldbe
estimatedfrom the dif'ference
in recalltime betweenoriginal
and reorderedsequences.
A baselineestimateof entry time
when therewas no memoryrequirementwas also obtained
by monitoringthe time requiredto copy the itemswhenthey
were displayedsimultaneously
on the screen.In rlrder to
obtain measuresof a more generalprocessingspeed,all
research participants also perfbrmed two computeradministered
versionsof the Dieit SymbolSubstitution
Test
( S a l t h o u s e1.9 9 2 ) .
METHOD
Subjects.- Characteristicsof the 38 young and 38 old
adults who participatedin this study are summarizedin
Table 2.
Procedure. - All subjectsperfbrmed the tasks in the
samefixed order:Digit Symbol, Digit Digit, Copy Digits,
Copy Letters,Memory Digits, and Memory Letters.Each
task was precededby a set of practicetrials, l8 eachin the
Digit Symboland Digit Digit tasks,and 6 eachin the other
tasks.The setof digit stimuliincludedthedigitsfrom I to 9,
andthe letterstimuliconsisted
of the first nineconsonanrs
in
thealphabet.(Vowelswereexcludedto avoidtheoccurence
of easilyremembered
words.)
Measuresof task-independent
processingspeedwere obtained fiom computer-administered
versions of the Digit
Symbol Substitutionand Digit Digit tasks (Salthouse,
1992).Thesetasksconsistedof the presentation
of eithera
digit anda symbol(Digit Symbol)or two digits(Digit Digit)
that the subjectwas to classifyas matchingor not matching
accordingto physicalidentity (Digit Digit), or accordingto a
codetablepresentedat the top of the display(Digit Symbol).
Becauseaccuracyaveraged)94o/a for young and old adults
P24l
TASK-SPECIFICPROCESSINGSPEED
in both tasks(cf. Table 2), the medianresponsetime across
90 trials, l0 repetitionsof eachdigit, servedas the primary
performancemeasurein eachtask.
All items for a given trial were presentedsimultaneously
on the computer screenin the copy tasks, and were to be
typed on the keyboard either in the original (left-to-right)
order of presentationor in a different order (alphabeticfor
letters,numericfor digits). The instructionaboutthe orderof
entry(i.e., the word "Original" or the word "Reordered")
was displayedsimultaneously,and timing fbr the first keystrokestartedwhen the items were displayed.
The items in the Memory tasks were presentedsequentially for I secondeach, with the prior item removed on
presentationof eachsuccessiveitem. The task was to recall
the itemseitherin the order in which they were presented,or
in a different order (alphabeticor numeric sequence).The
instruction about the order in which the items were to be
recalledwas displayedimmediatelyafter the last item in the
sequence,and timing of the keystrokeentry startedat that
moment.
Each combination of task (copy, memory), stimuli
(letters,digits), order (original, reordered),and number of
items(3, 5, or 7) was represented
by eight trials. The trial
typeswere blockedaccordingto task and stimulustype, but
the number of items presentedand the order of recall were
randomly mixed within blocks. All variablesexcept age
were manipulatedwithin subjects.
RESULTS
ANDDISCUSSIoN
Figure I displaysthe percentage
ofcorrectly recalledtrials
in eachmemorycondition.An Age x Stimuli x Number
x Order analysisof variance(ANOVA) was conductedon
the data summarizedin the Iigure. All main effects in this
a n a l y s i sw e r es i g n i f i c a n(tp < . 0 5 ) :A g e , F ( 1 , 7 4 ) : I l . 1 5 ,
M S " - 1 6 . 6 2 ;S t i m u l i ,F ( | , 1 4 ) : 2 5 2 . 4 2 ,M S " - 3 . 0 9 ;
Number, F(2,148) : 703.05, MS, : 2.62: and Order,
F(|,14) : 31.30, MS, : 5.03. Only two interactions
involvingage were significant(p < .05): Age x Number,
F(2,148) : 8.29, MS, : 2.62: and Age x Number x
S t i m u l i ,F ( 2 , 1 4 8 ): 5 . 4 1, M S , : I . 7 8 .I t i s n o t e w o r t htyh a t
the Age x Orderinteractionwas not significant,F(1 ,14) :
1 . 8 0 ,M S " : 5 . 0 3 ,p > . 1 5 , i n d i c a t i n gt h a tt h e o v e r a l la g e
differenceswere not significantlylargerwhen the processing
demands were increasedby requiring that the items be
recalled in a different order from that in which they were
presented.
The significantthree-wayinteractionwas decomposedby
conductingAge x Stimuli ANOVAs on the datawith each
numberof items.The Stimuluseffectwas significantin each
analysis(i.e., Fs > 10.2),but neitherthe Age main effect
nor the Ace x Stimulusinteractionwas sisnificantin the
T"bl" r. B".kg.""t
Srudy2
Study I
Young
old
Young
old
N
3rJ
38
3ri
38
%,Males
5-5.3
44.1
44.7
44.7
A g e ( r a n g e )2 0 . 2 ( 1 8 - 2 36) 9 . 7 ( 5 t l - 8 0 )2 0 . 1( 1 8 - 2 5 )6 8 . 4 ( 5 5 - 7 8 )
Education
(sD)
1 4 . 2( 1. 4 )
l s . 2( 2 . 8 )
1 3 . 9( 1 . 6 ) l s . O( 2 . 6 )
H e a l t(hS D ) | . 7 ( . 9 ,
l.tl (.8)
1.7 (.U)
1.7 (.7)
DigitSymbol
Accuracy
(sD)
9 5 2 ( 3 . 3 ) 9 4 . 5( 9 . - s ) 9 6 . 2( 2 . 2 ) 9 s . 4( 3 . 2 \
T i m e ( S D )1 . 1 2 6 ( 2 1 5 )l . t t 4 - 5 ( 3 6 3 )I , l 1 1 6 ( 2 2 6 I) . u 9 l ( 4 4 . s )
DigitDigit
Accuracy
(sD)
9 6 . 6( 2 . 2 ) 9 t t . 4( | . 7 )
9 1. 4 ( Z . O ) 9 8 . 0( 2 . 7 )
Time(SD/ 523(61)
767(213)
551(tt6)
768(202)
Norc. Education ref'ersto number of years of litrmal educationcompleted,and Healthref'ersto sell'-ratingof healthon a 5-point scalewith I :
excellentand 5 - poor. Digit Symbol and Digit Digit accuracyvaluesare
percentagecorrectresponses,and time valuesare in msec.
Letters
Digits
100
Youno - Orioinal
o
o
o0
+
Youno- Re€rdered
-aOld -Odginal
o7s
Old - Re0rdered
(t
75
G
L
F
.*
(J
5O
50
o
o,
(d
5
zs
o
25
o
(L
5
7
Numberof ltems
5
7
Numberof ltems
Figure l. Mean percentage
c o r r e c ta c r o s so r i g i n a l a n d r e o r d e r e ds e q u e n c e s f o r 3 , 5 , a n d T i t e m s f o r y o u n g a n d o l d a d u l t s , S t u d yl .
P248
SALTHOUSEAND COON
databasedon 3 items, and only the main effect of Age was
significantwith 5 items,F(l,74) : 8.31, MS. : 5.00. Both
the main effectof Age, F( I ,'/4) : 22.56,MS": 3.06, and
the Age x Stimuli interaction,F(\,14) : 8.86, MS" :
1.22, weresignificantwith 7-itemsequences.
Inspectionof
Figure I revealsthat this latter interactionis attributableto a
largeragedifferencewith digits thanwith letters,but the low
level of performancewith lettersraisesthe possibility that
the interactionis an artifactof a measurementfloor with the
letter stimuli.
A detailedexaminationwas also conductedof processing
times obtainedduring the performanceof the task. Very few
subjectswere accurateon many trials involving 7 items, and
thus the analyseswere restrictedto 5-item sequences.The
analyseswere further limited to subjectswith at least one
correcttrial in both the original and the reorderedsequence
with eachtype of material, in order to ensurecompletedata
from eachsubject.This resultedin subsamplesof 29 young
adultsand 25 older adults.The agesof the subjectsin the
subsamplewere similar to thosein the entire sample(i.e.,
young 20. I vs 20.2, and old 68.6 vs 69.7), alrhoughboth
age groups in the subsamplehad somewhat faster Digit
Symbol scoresthanthe total sample.That is, the meanswere
1,079 msec in the subsamplevs 1,126 msec in the total
samplefor young adults, and 1,792 msec in the subsample
vs 1,845msecin the total samplefor older adults.
The meansacrosssubjectsof the mediantime per item in
correctmemory trials and correct copy trials for each serial
positionaredisplayedin Figure2. SeparateAge x Order x
Serial PositionANOVAs conductedon the copy and memory data with digit and letter stimuli revealedthat all main
effectsand interactionswere significant(p < .05), exceptfor
the Order effect in the Digit Copy condition. The data in
Figure 2 indicatethat the first entry is ratherslow, and that
the remaining items are enteredquite rapidly except when
the items must be reorderedfrom memory, and particularly
with letterstimuli. Both young and old adultswere somewhat slowerin the recallof the third item, possiblybecause
the items were initially reordered in two-item groups.
(Resultsfrom a similar analysisbasedon the datafrom l6
Digits
Letters
MEMORYTIME
MEMORY
TIME
o
I
o
tr
->
(D
E
tr
o
o
c
o
oa
o
IE
c
.g
&-'
sux)
2@O
i@o
o
Young - Original
+
Young - Re-Ordered
-O-
Old - Original
Digits
Eooo
COPYTIME
7m
Letters
-F
Old - ReOrdored
COPYTIME
-O-
E
(t)
20m
1()oo
1
2
3
4
SerialPosition
s
o
1
2
3
4
SerialPosition
Figure 2. Median responsetime in 5-item copy and memory trials with original and reorderedsequences
fbr young and old actults,Stu6y I
TASK-SPECIFICPROCESSING
SPEED
young subjectswith at leastone correct trial on the 7-item
lists in every condition were consistentwith this interpretation becausethere were two peaks in the recall times of 7letterlists, one at the third item and anotherat the fifth item.)
The slower entry on the third item is largely restrictedto
memoryprocessing,becauseit is not pronouncedin the copy
task. Perhapsthe most importantpoint to be noted from the
data in Figure 2 is that, although the older adults were
generallyslower than the young adults, the overall pattern
appearsvery similarin the two groups.
The median entry times in the copy and memory tasks
with original and reorderedrecall were usedas predictorsof
recall accuracywith 5-item lists. (Data from 3-item and 7item lists were not includedto avoid measurementceilings
and floors.) An additionalpredictor was a compositemeaprocessingspeedcreatedby aversure of task-independent
aging the z-scoresfor the time measuresfrom the Digit
Symboland Digit Digit tasks,which hasa correlationof .71
with each other. The resultsof the hierarchicalregression
analysesfor the 54 subjectswith completedata are summarized in Table 3. Regressionanalysesincludingthe taskspecific speedmeasureswere designedsuch that the times
for the simplestor most fundamentalprocesses
were entered
first, followed by times representingmore complex processes.That is, the entry time in the copying condition
precededthe entry time in the memory condition in the
prediction of both original and reorderedrecall accuracy,
and the entry time in original recall precededthe entry time
in reordered recall in the prediction of reordered recall
accuracy. Analyses were also conductedwith component
durations derived by subtractingone time from another
(e.g., entrytime with recallin originalorderwas subtracted
from entry time in reorderedrecall to yield a measureof
reorderingtime). However, the resultsof thoseanalysesare
not reported becausethese difference score measureshad
low reliability and exhibited very weak relations to the
criterion memory performancemeasures.
Five major points should be noted about the results in
Table 3. First, there is no significantage-relatedvariancein
the accuracyof recalling 5 items in their original order of
(Equationl). Second,thereis significantagepresentation
related variance in the accuracy of recalling items in a
different order (Equation5), and it is still significantafter
recall accuracyin the original order is controlled(Equation
7). Third, both the age-relatedvariance(Equation6) and the
influence of recall in the original order (Equation 8) are
reducedwhen the task-independentspeedmeasureis controlled. In fact, examinationof Equations6 and 8 reveals
that there was no residualage-relatedvariancein the measureof reorderedrecall after eliminatingthe variationin the
task-independent
speedmeasure.
The fourth point to note from Table 3 is that significant
relationswere evidentbetweenthe task-specificspeedmeasuresand memory performance(Equations3, 9, and l0).
Furthermore, the age-relatedvariance in reorderedrecall
was substantiallyreducedafter controlling the variation in
these task-specificmeasures(Equations9 and l0). However, the fifth point to be noted is that the reduction in the
age-relatedvariancewas greater,and the relationsbetween
memory and the task-specificspeedmeasuresweaker, after
P249
the task-independent
speedmeasurewas controlled (Equat i o n sl 1 a n d l 2 ) .
The relation of age on the task-specificprocessingdurations before and after control of the task-independent
speed
index was also examined in regressionanalyses,with the
resultssummarizedin Table 4. Only in recall time for letters
in their original order (Equation6 with letterstimuli) was the
age-relatedvariancesignificantlygreaterthan zero after the
task-independentspeedindex was controlled. Even in this
case,however, the degreeof attenuationafter control of the
varianceassociatedwith the task-independentspeedmea-
Table3. Prediction
of Performance
With 5-ltemMemory
Lists(N : 54),StudyI
Equation
Original
I
Digits
Letters
Cum. Incr. F for
R2
R2 lncr. R2
Cum. Incr. F fbr
R2
R2 Incr. R2
Age
.019 .019
2
Speed
Age
.052
.064
3
CopyTime-O .096 .096 5.42*
MemTime-O .l l5 . 0 1 9 t . t 2
Age
. l 1 7 .002
.06
4
Speed
CopyTime-O
MemTime-O
Age
Reordered
5
Age
1.03
.052 2.82
.0t2
.69
.052 .052 2.tt4
.053 .0-s3 2.86
.056 .003
. l8
.099 .099 5.47*
.099 .000
.00
.099 .000
.03
.052
.096
.l 16
.123
.052 2.89
.044 2.46
.020 I .16
.007
.39
.053
.099
.099
.099
.124
.053 2.titt
.046 2.5|
.000
.00
.000
.01
.124 7.39*
.t94 .t94 12.494
6
Speed
Age
.150 .150 9.05*
. 1 5 2 .002
.08
.221 .227 15.05*
.230 .003
.21
7
Original
Age
.100
.197
.100 6.32*
.091 6. 17+
.t85 .185 13.66x
. 3 0 9 . t 2 4 9 . tl *
8
Speed
Original
Age
. 1 5 0 . 1 5 0 9 . 5 1 + . 2 2 7 . 2 2 7 1 7 |. 3 *
. 2 0 5 .055 3.44
. 3 3 6 . 1 0 9 t i .1 9 +
.209 .004
.25
.33'7 .001
.l r
9
CopyTime-R . 183 . 1 8 3 I l . 7 l *
MemTime-R .205 .022 t.42
Age
.219 . 0 1 4
.89
.113 .113 I 1.84*
.178 .005 .34
.268 .090 6.09*
l0
MemTime-O .250 .250 t6.74+
M e m T i m e - R . 2 5 2 .002
.09
Age
.253 .001
.09
.148 .148 9.46*
.148 .000
.00
.218 .070 4.46*
ll
Speed
CopyTime-R
MemTime-R
Age
.150
.199
.214
.219
. 150 9.44*
.o49 3.01
.015 .99
.005
.30
.227
.256
.2't4
.28|
.22'7 15.48*
.029 1.98
.Oltt t.2l
.007
.50
12
Speed
MemTime-O
MemTime-R
Age
.150
.254
.255
.255
.150 9.90*
.104 6.83*
.001
.06
.000
.01
.227
.248
.258
.258
.227 15.00*
.021 1.35
.010 .66
.000
.01
Note. Cum. : cumulative;lncr. : increment;CopyTime-O : median
entry time in copy condition with original sequence;CopyTime-R :
medianentry timp in copy conditionwith reorderedsequence;MemTime-O
: medianentry time in memory condition with original sequence;MemTime-R : median entry time in memory condition with reordered
sequence.
*p < .05.
P250
SALTHOUSEANDCOON
Table4. Prediction
of Recall(or Entry)TimeMeasures
With 5-ltemListsfromAge andSpeed
(N : 54),SrudyI
Digits
Cum.
R2
Equation
Copy-Original
I
Age
2
Speed
Age
Copy-Reordered
3
Age
4
Speed
Age
Mernory-0riginal
-5
Agc
6
Speed
Age
Speed
A8e
Letters
F lbr
lncr. R2
Cum
R2
Incr.
R2
F fbr
lncr. R2
.409
.409 36.06*
.446
.446 4 1 . 8 8 *
.554
.-554
.554 63.28*
.000
.00
.465
.490
.465 16.52+
.025 2.46
.2U9
. 2 8 9 2 1. 1 4 *
.254
.254
.487
.503
.487 49.99*
.016 1.68
.340
.340
.340 26.28*
.000
.00
MprHon
.421
.421 37.86+
.422
.422 37.119*
164
.419
.464 45.441.
.01-5
| .43
. 3 1|
. 3 tI
.422 .l I I
Subjects.- Characteristics
of the subjectsamples,38
adultsin eachof two agegroups,aresummarized
in Table2.
None of theseindividualshad participated
in Study I .
Menrory-Reordered
7
Age
.200
tt
Incr
R2
The present study therefore included manipulationsof
stimuluspresentationtime to determinethe relationbetween
a measureof encodingtime and recall accuracy.The stimulus durationrequiredto achievea specifiedlevel of accuracy
was then usedas the rneasureof the individual'sstimulus
encodingtime. In most other respectsthe study was similar
to Study l. That is, the numberof subjectswas the same,
they were recruited in the same manner, and the Digit
Symbol and Digit Digit taskswere again usedto provide an
index of task-independent
processingspeed.However,becauseof the time neededto examinethe effect of variations
in stimuluspresentation
time, only the memorytaskwith the
5-itemlettersequences
was administered.
.215
.775
.200
| 3.001,
.275 19.35+
.000
.01
17.73+
2l .46r,
9.70*
.214 .2t4
l4.lz+
.lt{l
.2| tt
t2.29+
I.96
.luu
.030
C u m . : c u m u l a t i v c ;I n c r . : r n c r e m e n t .
'ip < .05.
s u r e w a s c o n s i d e r a b l(ei . e . , l 4 o / o , f i o m a n R ' o f . 4 2 2 t o
.I I l ) .
To summarize,the resultsof this studv reolicatedthe
linding of substantialsharedvarian." urnnngage, perceptual
comparisonspeed,and memory perfbrmance.In addition,
the resultsindicatedthat althoughsome of the task-specific
speedmeasureswere significantlyrelatedto the measuresof
memory perfbrmance,only a small proportionof thoserelations was independentof a presumablymore generalspeed
indexderivedf'romseparate
tasks.Moreover,nearlyall of the
age-relatedvariancein the task-specificspeedmeasureswas
sharedwith the task-independent
speedmeasure.
Study 2
Kliegl and colleagues
haverecentlyprovidedconvincing
evidencethat input processingtime is an importantfactor in
the age diff'erencesin serial recall by the finding that older
adults need more time to achievethe same level of recall
performanceas young adults (Kliegl, Smith, & Baltes,
1989;Thompson
& Kliegl, 1991).Furthermore,Thompson
and Kliegl (1991)cite an unpublishedstudy by Kliegl and
Lindenbergeras having found that the "age diff'erencesin
recall accuracywere completelyaccountedfor by rndividual
differencesin criterion-referenced
encodins times" (o.
5 4 4 ) .W h a ti s n o tc l e a rl r o m t h eK l i e g ls t u d i e s l w
s h e t h etrh e
influenceof input processingtime is independentof a more
general,or task-independent,
processing
speed,andwhether
a similar relationbetweenencodingtime and recall performanceexistsamonguntrainedsubjects(i.e., research
participantsin the Kliegl studiesreceivedextensivetrainingwith
the method-of-locimnemonicstrategy).
Procedure.- The taskswere perfbrrnedin the fbllowing
orderby all subjects:Digit Syrnbol,Digit Digit, and Memi
ory Letters.The Digit Symbol and Digit Digit taskswere
identicalto thoseof Study I . Mean levelsof accuracywere
)95o/oin both tasksand both agegroups(cf. Table 2), and
thus mediantime per response
servedas the primarydependentvariablein thesetasks.
The memory task dif-fbredfrom that in Study I by using
only 5-letterstimulussequences
and by rcplacingthe fixed
stimulus duration of I second/itemwith a duration that
variedacrosstrial blocks.A difl'erentrangeof presentation
times was used for young and old adults, becausepilot
research
revealedthatthe olderadultswereseldomsuccessful with presentation
durationsof 100 msec.Young adults
therefbrereceiveddurationsof 100, 200, 400, 800, and
I ,600msec/item,whereasolderadultsreceiveddurationsof
200, 400, 800, 1,600,and 3,200 msec/irem.Blocksof l6
trials each were presentedfiom the slowest to the f'astest
durations,and then againin the reverseorder suchthat each
presentation
durationwas represented
by 32 trials.All trials
in a given block had the samepresentation
time, but on a
randomlyselectedhalf of the trials within each block the
letterswereto be recalledin theoriginalorder,andon half of
the trialsthey were to be recalledin alphabeticorder.
Rp,sulrsANDDISCUSSToN
The meanpercentageof 5-lettertrials recalledcorrectlyas
a functionof stimuluspresentation
durationis displayedin
Figure 3. An Age x Order X Duration ANOVA was
conductedon the accuracyfor trials with presentationdurationscommonto both agegroups(i.e., 200, 400, 800, and
1,600 msec).All threemain effectswere significant(p <
. 0 5 ) ;A g e ,F ( 1 , 1 4 ): 4 6 . 1 8 ,M S " : 5 7 . 7 1 O
; rderF
, (I,14)
: 21.'17,MS" - 7.59;and Duration,F(3,222) : 146.73,
MS" : 6.03, but noneof the interactions
involvingage was
s i g n i f i c a n(ti . e . , a l l F < 1 . 3 ) .
Two methods were used to derive estimatesof each
individual'sinput processing
time or task-specific
encoding
duration. One method was basedon the fact that the means
summarizedin Figure 3 could be accurately(i.e., all R'
TASK-SPECIFICPROCESSINGSPEED
Youno - Orioinal
--'
- R6ordqed
- ra- -.
Old - Original
Old - Reords€d
--€--.
o
o 6 0
(J
o
6
c
o 4 0
o
o
&
1m 'X,"""n,iilon
o",,ff-t'l$
32m
Figure 3. Mean percentagecorrcct fbr original and reorderedS_letter
sequencesas a function of presentationdurationfor ytrungand old adults.
S t u d y2 .
>.99) describedby negativeexponentialfunctionsof the
fbrm:
P r o p o r t i oC
n o r r e c t: C * ( l - e x p ( - ( t i m e * A ) * B ) ) .
The C parameterin this equationcorresponds
to the asymptotic levelof accuracy,theA parameter
to the time at which
accuracybeginsto increase,and the B parameterto the rate
of increasein accuracywith additionaltime. Analysesbased
on the group meansrevealedthat relativeto older adults,
young adultshad smallerA parameters
(i.e., .086 and .065
for originaland reorderedrecallfbr young adults,and .100
and . l5l for originaland reorderedrecallfor older adults),
larger 8 parameters(i.e., 1.19 and 5.28 for original and
reorderedrecall fbr young adults, and 3.63 and 4.73 for
originaland reorderedrecallfor older adults),and largerC
parameters(i.e., .7ll and .689 fbr original and reordered
recall for young adults, and .520 and .428 for the correspondingvaluesfbr older adults).Becausethe asymptotic
level of accuracywas only .428 with reorderedrecall in the
older group, the time correspondingto an accuracyof .25
(i.e., correctrecallof 25o/o
ofthe lists)servedas the measure
of encodingtime. (SeeAppendix,Note I .) The timesbased
on the equationsfrom the group meanswere 142 msec for
original recall in young adults, 150msecfbr reorderedrecall
in youngadults,28 I msecfor originalrecallin olderadults,
and 337 msecfor reorderedrecall in older adults.
Becausethe goal was to derive encodingtime estimates
from eachsubject,attemptswere madeto fit negativeexponential functionsto the original and reorderedrecall data of
eachsubject.Data of somesubjectscould not be fit because
of small or unsystematicaccuracy variations, and time
estimatescould not be derived in five additional cases
becausethe accuracyasymptoteswere lessthanthe criterion
valueof .25. Age x Order ANOVAs wereconductedon the
parameters,and on the estimatedencoding times correspondingto an accuracyof.25, for the 32 youngadultsand
27 older adults with relevant data. No effects were significant in the analysesof theA, B, or R2parameters,and only
the Age effect was significant(p < .05) on the C parameter,
P25r
F(1,57) = 22.83, MS. : .037. All three effecrs were
significant(p < .05) in the analysisof estimatedencoding
t i m e :A g e , F ( 1 , 5 7 ) : 1 0 . 3 3 ,M S " - . 1 3 3 ;O r d e r ,F ( I , 5 7 )
: 13.52,andAge x Order,F(l ,57) : 9.69, MS, : .O4t.
Means of the estimatedtimes were 169 msec for young
adultswith recall in the original order, 195 msecfor young
adults with recall in alphabeticorder, 355 msec for older
adults with recall in the original order, and 512 msec for
older adultswith recall in alphabeticorder.
The second method used to estimateeach individual's
task-specificencodingtime was basedon linear regression
equationsrelating log presentationtime to accuracyacross
thethreeshortestdurations(i.e., 100,200. and400 msecfor
young adults,and 200, 400, and 800 msecfor older adults).
These equations were then used to predict the duration
requiredto recall 25o/ootithe lists correctly. Data from five
subjects,one young and four old, are omitted from these
analysesbecausethe encodingduration estimateswere not
meaningfuldue to correctrecall of 125o/oof the lists across
all threeof the shortestpresentationdurations.Meansof the
correlationsrepresentingthe goodness-of-fitof the regression equationswere .92 and.90 for original and reordered
trials,respectively,
for the 37 youngadults,and .8 I and .80
for the two typesof trials for the 34 older adults.An Age x
Order ANOVA on the correlationsrevealedthat only the
A g e e f f b c tw a ss i g n i f i c a n F
t ,( I , 6 9 ) : 7 . 3 2 ,M S . : . 0 5 1, p
< .05, indicatingthatthe relationsbetweenstimuluspresentation time and accuracywere somewhatlessprecisefbr the
older adults.The predicteddurationscorrespondin
g to 25o/o
accuracyaveraged139msecfbr recallin the originalorder
and 160 msecfbr recall in alphabeticorder for the 37 young
adults,and 412msecand 626 msec,respectively,
fbr the 34
olderadults.Only the Age maineffectwas significantin the
A g e x O r d e r A N O V A , F ( 1 , 6 9 ) : 2 4 . 0 3 ,M S , : . 4 2 .
Although the patternof meanssuggeststhat an interaction
might be present,the Age x Order interactionwas not
significant,ic'(1,69)< 1.0.
Correlationscomputedbetweenthe predictedtimes correspondingto correctrecallof 25o/oof the trials derivedfrom
the two methods were very high, with r : .95 for the
estimatesin the original order condition and r : .86 for
estimatesin the reorderedcondition. Becauseestimateswere
available for more subjects when the linear regression
methodwas used(i.e., 7l comparedwith 59), subsequent
analyseswere basedon the encodingtimes derivedfrom the
linear regressionprocedure.It should be noted, however,
that the high conelationsbetweenthe two setsof estimates
suggestthat very similar resultswould have been obtained
with encoding times derived from negative exponential
functions.
Figure 4 displays the mean of the median recall times
(averagedacrosstrials with 1,600and 800 msecpresentation
durations)by serial position. (Thesepresentationdurations
were selectedbecausethey were the largestdurationscommon to both agegroups.)An Age x Order x Serialposition
ANOVA revealedthatall main effectsandinteractionsexcept
for the three-wayinteractionweresignificant(F's > 3.9, p <
.05). Inspectionof Figure 4 revealsthat the patternis very
similar to the top right panel of Figure 2, in that the older
adultswere always slower than the young adults, and recall
P252
SALTHOUSEAND COON
8000
Youm - Odoinel
Table5. Prediction
of Performance
With s-ltemMemoryI_ists
From 1,600and800MsecPresentation
DurationTrials
( N : 7 l ) . S r u d v2
+--
ao
a
g
Yourc- --- RFordered
a--
7000
Old -gdSinal
Old - Reordsed
--a--
6(no
o
.E
F
o
a
so@
Original
I
5 0 m
a
e
Incr.
R2
F for
lncr. R2
3ooo
Age
.279
.279
26.72*
2
Speed
Age
.t67
.280
.t67
.lt3
15.75*
10.69*
3
EncodingTime
Age
. 3 0I
.4t2
.301
.lll
34.84E
I2.88*
4
MemTime-O
Age
.326
.365
.326
.039
34.88*
4.25*
5
Speed
EncodingTime
Age
.t67
.376
.415
.t6t
.209
.039
19.102*
24.02*
4.42*
6
Speed
MemTime-O
Age
.167
.326
.373
.167
.159
.047
17.828
17.06*
4.99*
Speed
EncodingTime
MemTime-O
Age
.t67
.376
.432
.453
.167
.209
.056
.02t
20.124
25.3I *
6.67*
c
.q
E
Cum.
Equation
20@
1
2
s
serial losition
Figure 4. Median responsetime in 5-letter memory trials with original
for young and old adults, Study 2.
and reorderedsequences
time for reorderedsequenceswas slower than recall in the
original sequence,particularlyin the third serialposition.
Table 5 containsresultsof the analysesparallelingthose
reportedin Table 3 for recall with presentationdurationsof
800 and 1,600msec,which most closelycorrespondto the
l-sec presentation
durationusedin Study l. Severalpoints
shouldbe notedabouttheentriesin this table.First,the agerelatedeffects in letter recall were much larger in this study
than in the previous study, both for recall in the original
o r d e r ( i . e . , R ' f o r a g e o f . 2 7 9 v s . 0 5 2 ) a n d f o r r e c a l li n
a l p h a b e t i co r d e r ( i . e . , R 2 f o r a g e o f . 3 8 1 v s . 1 9 4 ) . T h e
reasonsfor the larger age relations in this study compared
with thosein Study I are not obviousbecausethe procedure
was generallysimilar and the samplesappearcomparablein
most respects(seeTable 2). The fact that more trials were
administeredin the relevantconditionsin this study compared with the previous one (i.e., 32 vs 8) may have
contributedto the differenceacrossstudies,eitherbecauseof
differential practice effects or increasedreliability of the
memory measure.(See Appendix, Note 2.) Second, although control of the compositeperceptualspeedmeasure
substantiallyreducedthe age-relatedvariance(i.e., percentage attenuationsof 597oand,737ofor original and reordered
recall, respectively), the residual age-related variance
(Equations2 and 9) was significantly > 0. This indicates
that, unlike Study l, control of the task-independent
speed
index is not sufficientto accountfor all of the significantagerelated variance in immediate recall in this study. Third,
significantage-relatedvarianceremains in the measureof
reorderedrecall after the measureof recall in the original
order is controlled (Equation l0), but not after the taskindependentspeedmeasureis also controlled(Equation I l).
Both of these findings are consistentwith the results of
Study l.
Another important finding to note from Table 5 is that
statisticalcontrol of the task-specific(i.e., Encoding Time
and Recall Time; Equations3, 4, 12, and 13) speedmeasuresresulted in substantialattenuationof the age-related
variancein memory performance.That is, the percentage
attenuation of the age-relatedvariance was 6O.2Voand
'1
Reordered
E
2.58*
Age
. 3 8|
.38|
42.46*
9
Speed
Age
.307
.403
.307
.l0l
34.964
10.88*
l0
Original
Age
.5t2
. 5 9|
.5t2
.079
8 5 .1 2 *
13.21*
ll
Speed
Original
Age
.307
.594
.609
.30'7
.287
.015
52.57*
49.t4*
2.4' 7
t2
EncodingTime
Age
.590
.661
.590
.011
120.52*
t5.88*
I3
MemTime-O
MemTime-R
Age
.37s
. 4t 8
-+t-)
.375
.043
.055
41.65*
5.49*
7.08*
l4
Speed
EncodingTime
Age
.307
.644
.67|
.30'7
.33'7
.027
62.61*
68.62+
5.66*
l5
Speed
MemTime-O
MemTime-R
Age
.307
.410
.431
.473
.307
.103
.02t
.o42
38.49*
12.89*
2.68
5.27*
l6
Speed
EncodingTime
MemTime-O
MemTime-R
Age
.307
.644
.657
.658
.678
.307
.337
.013
.001
.020
61.93*
6'7.87*
2.64
.2'r
3.96
Nole. Cum. : cumulative; Incr. : increment; Encoding Time :
predicted stimulus presentation time required to achieve 25Eo accuracy;
MemTime-O : median entry time in memory condition with original
sequence;MemTime-R : median entry time in memory condition with
reordered sequence.
*p < .05.
86.07o after control of the encoding time and recall time
speedmeasuresfor recall in the original order, and 19.87o
and85.6Vo, respectively,after control of thesemeasuresfor
P253
TASK.SP
ECIFIC PROCESSINGSPEED
recall in alphabeticorder. The residualage-relatedvariance
was significantly greaterthan zero, indicating that factors
independentof both thesespeedmeasurescontributedto the
age-relateddifferences in memory performance. In this
respect,the current resultsdiffer from those of Kliegl and
Lindenberger,as reportedin Thompsonand Kliegl (1991).
Another noteworthyresult from Table 5 is that the relations
betweenthe task-specificspeedmeasuresand memory were
smaller after control of the composite task-independent
s p e e dm e a s u r (ei . e . , E q u a t i o n3s v s 5 , 4 v s 6 , l 2 v s 1 4 ,a n d
l3 vs l5). This suggests
that only a portionof the variance
sharedbetween memory and the task-specificspeedmeasureswas distinct from the presumablymore general,taskindependentspeedmeasure.
A final setof regressionanalyseswere similar to thoseof
Table 4 in that they examinedthe extent to which the agerelatedvariancein the task-specificspeedmeasurescould be
accounted for by the task-independentspeed measure.
Resultsof theseanalysesare presentedin Table 6. Notice
that only with recall time for reorderedrecall (Equation8)
was therenot significantresidualage-relatedvariancein the
task-specificmeasuresafter control of the presumablymore
generalmeasuresof perceptualspeed.However, it should
also be noted that each of the task-specificspeedmeasures
had considerableage-relatedvariancein common with the
perceptual speed measure because control of the taskindependent speed measure resulted in attenuationsof
39.2o/o,78.la/o,86.5o/o,ancl94.4o/o,respectively,for the
measuresof original encodingtime, reorderedencoding
time, originalrecalltime, and reorderedrecalltime.
Times
Table6. Prediction
of Task-Specific
Processing
(N : 7l), Study2
FromAgeandSpeed
Cum
p2
Equation
In c r
R2
F fbr
lncr. R2
EncodingTime Measure
Original
I
2
Reordered
3
4
.166
.166
13.74*
Speed
Age
.06tt
.t69
.068
.l0t
5.58*
8.29*
Age
.237
237
2 1. 4 1 +
Speed
Age
.204
.256
204
052
18.64*
4 . 74 *
Age
.443
.443
54.17+
Speed
Age
.462
.462
.060
65.12't'
tl.'10*
A8e
.33u
.437
.456
. 33 8
+
35.23
.437
. 0l 9
-54.6-5*
2.45i'
Recall Time Measure
Original
5
6
Reordered
7
tJ
Speed
Age
.) ll
*p < .05
General
Processing
Speed
General Discussion
The primaryquestionof interestin this projectwas, how
does a slower processingspeed contribute to age-related
differencesin memory'?The researchstrategyconsistedof
trying to measurethe durationsof componentspostulatedto
be involvedin a particularmemorytask,andthenexamining
how thesetask-specificspeedmeasureswere relatedto age
and to performancein the memorytask, and how both setsof
relationswere influencedby speedmeasuresderived from
independenttasks.
Figure 5 illustratesa frameworkwithin which the present
researchcan be interpreted.A major purposeof the studies
reportedhereinwas to examinethe plausibilityof eachof the
relationsrepresentedin this figure.
Three measuresof task-specificprocessingspeed were
examinedin thesestudies.The speedof initial registration
and encodingof the items was representedby the stimulus
presentationduration neededto correctly recall a specified
percentage
of lists,the speedofaccessingor retrievingitems
from memory was representedby the additional time required to enter items from memory comparedwith entering
t h e m f r o m t h e d i s p l a y( i . e . , E q u a t i o n s3 , 4 , 9 , a n d l l i n
Table 3), and the speedof reorderingthe items was representedby the additional time to enter items in a different
order relative to entering them in the original order (i.e.,
Equationsl0 and I 2 in Table 3 , and EquationsI 3 and 15 in
Table 4). The measuresof a presumably more general
processingspeedwere derived from independenttasks re-
Memory
Performance
Task-Specific
Processing
Speed
Figure 5. lllustration of hypothesizedrelations among age, memory
performance,and generaland specilicmeasuresof processingspeed.
quiringdecisionsaboutthe physicalidentityor associational
equivalenceof a pair of items. Previousresearch(e.g.,
Salthouse,1992)hassuggestedthat memoryfactorscontribute very little to the age-relatedvariationin thesetasks,and
that performanceon the taskscan be interpretedas primarily
reflectingthe speedwith which relativelysimple perceptual
and cognitiveoperationscan be executed.
Although path analytic or structural equation methods
were not used in this project becauseof the small sample
sizes, results from the curent studiesnonethelessallow a
number of inferencesto be drawn concerningthe relations
in Figure5. First, considerthe pathslinking age
represented
to memory performanceeither directly (Path I ) or indirectly
by means of a reduction in general, or task-independent,
processingspeed (Paths 2 and 5). (Note that a direct influencein this contextmerely meansthat it is independentof
P254
SALTHOUSE
AND COON
the other variables being considered, and not that it is
unmediatedby any factor.) The substantialreductionin the
age-relatedvariance in memory after the variation in the
index of general processingspeedwas controlled implies
that Paths2 and 5 are involved in the relationsbetweenage
and memory. However, the discoverythat significantresidual age-relatedvarianceremainedin Study 2 after control of
the index of generalspeedsuggeststhat Path I sometimes
contributes to the age-memory relations. Both of these
pathways can be viewed as representingunexplainedinfluences,becausethe mechanismsby which increasedage,
or a slower speed of performing perceptualcomparison
tasks, contributesto lower levels of memory performance
cannotyet be specified.
Second, similar types of comparisonswith the taskspecificspeedmeasuressuggestthat Paths3 and 6 in Figure
5 arealso involvedin the relationsbetweenageand memory.
That is, becausethe age-relatedvariance in memory was
reducedwhen the task-specificspeedmeasureswere controlled, it can be inferredthat someof the agedifferencesin
memory are mediatedthrough reductionsin the speedof
memory-specificprocessing.The causalmechanismslinking these task-specificspeedmeasuresto memory performancemay be relatedto impairedquality of encoding,or to
a greaterloss of information.That is, the degreeof elaboration of encodedinformationmight be less,or the proportion
of displaced or decayed information greater, when more
time is requiredto encode,access,or process(i.e., reorder)
to-be-recalledinformation.
And third, informationabout the relationsbetweenthe
general and specific measuresof speed is available from
results of regressionequations predicting specific speed
from age and the generalspeedmeasure(Tables4 and 6),
and predicting memory perfbrmancefiom age, the general
speedindex, and the specificspeedmeasures(Tables3 and
5). The formerequationsarerelevantto the pathslabeled2,
3, and4 because
the strengthof the indirectinfluenceof age
on the specific speed measures(Paths 2 and 4) can be
ihferred to be proportionalto the degreeto which the agerelatedvariancein the task-specificspeedmeasuresis attenuatedafter control of the more generalspeedmeasure.The
resultsin Tables4 and 6 indicatethat much of the relation
betweenage and the task-specificspeedmeasuresis mediatedthroughgeneralspeed(via Paths2 and 4), althoughthe
existenceof significantindependentage-relatedvariancein
some measuresindicatesthat Path 3 can also contributeto
the relation.
Sequentialexaminationof the influence of the general
speedindex, the task-specificspeedmeasures,and age on
the measuresof memory performanceis informative about
the Pathslabeledl, 5, and 6 in Figure5. Althoughthereis
some variation acrossthe task-specificspeedmeasures,it
appearsthat all of the paths illustrated in Figure 5 can
contributeto the relationsbetweenageand memory. That is,
the age-relatedvariancein memory performanceis reduced
when either the generalor the specific speedmeasuresare
controlled, and, at least in some cases, there is still a
significant relation between memory performanceand the
specificspeedmeasureeven afterthe generalspeedmeasure
was controlled. Analyses with the general speedmeasure
enteredin the prediction equation after the specific speed
measureswere not reportedin Tables 3 and 5, but results
from these analyseswere consistent with the preceding
interpretation.That is, in at least severalcombinationsof
predictorsand criterion memory measures,the influenceof
the generalspeedmeasurewas still significantafter removing the varianceassociated
with the speci{icspeedmeasures.
This occurredafteralteringthe orderof entry of the variables
in Equations1l and 12 with letterstimuliin Table 3, and in
E q u a t i o n5s , 1 4 , 1 5 ,a n d l 6 i n T a b l e5 .
To summarize.the resultsof thesestudiesconfirm earlier
findingsofa large influenceofprocessingspeedon adult age
differencesin memory. Earlier researchis also extendedby
revealingthat increasedage is associatedwith slower processingof severalcomponentshypothesizedto be involved
in the performanceof theseparticularmemory tasks.However, it is important to note that the task-independentand
task-specificmeasuresof speedhave substantialage-related
variancein common, and thus they are not independent.
Indeed,the patternof resultsis consistentwith the interpretation that the age differencesin the specificspeedmeasures,
includingthe encodingtime measurebasedon procedures
similarto thoseusedby Kliegl andcolleagues
(Kliegl et al.,
1 9 8 9 ; T h o m p s o&n K l i e g l , l 9 9 l ) , a r ep a r t i a l l yc a u s e db y a
moregeneralage-related
slowing.
Althoughthesestudiesfocusedon encodingtime, memory retrieval or accesstime, and processingor reordering
time, it seemsreasonable
to expectsimilarresultswith other
measures
of relevantprocessing
durations.For example,the
reports that young and old adults differ in both memory
performance and in subvocal rehearsal rate (Salthouse,
1980)or articulationrate (Kynette, Kemper, Norman, &
Cheung,1990)couldbe viewedas additionalmanif'estations
of the interrelations
of age, a relativelygeneralprocessing
speed,and memoryexaminedin thesestudies.Moreover,
the current perspectiveleads to the expectationthat if the
durationsofcomponentssuchasassociation,
elaboration,
or
organizationcould be assessed,
then they would also be
relatedto age and to more generalmeasuresof processing
speed, as well as to memory. The causesof age-related
differencesin the postulatedgeneralprocessingspeedhave
not yet been determined,and the exact manner by which
slowerprocessingreducesmemory effectivenessstill cannot
be specified.Nevertheless,
the resultsof thesestudiessuggest that at least some of the age-relatedimpairments in
memory are attributableto declineswith age in the speedof
relevantprocessing,and that the speedof memory-specific
processingis closely related to the speed with which the
individual can perform other relatively simple tasks.
AcKNowLEDcMENTS
This researchwas supportedby a grant from the National Instituteon
Aging (R37-AG06ll26).
We thank Philip Allen, Nelson Cowan, Robert Kail, and Reinhold
Kliegl for helpful commentson an earlier versionof the manuscript.
Addresscorrespondence
to Dr. Timothy A. Salthouse,School of Psychology, Georgia Instituteof Technology,Atlanta, GA 30332-0170.
REFERENcES
Babcock,R. L., & Salthouse,T. A. ( I 990). Effectsof increasedprocessing
P255
TASK-SP
ECI FIC PROCESSI
NG SPEED
demands on age differences in working memory. Psychologl tuttl
Aging,5,421 428.
G i c k , M . L . , C r a i k . F . l . M . , & M o r r i s ,R . G . ( 1 9 8 8 ) .T a s kc o m p l e x i t ya n d
age differencesin working memory. Memory'& Col1rtition,16, 353361.
K l i e g l , R . , S m i t h , J . , & B a l t e s ,P . B . ( 1 9 8 9 ) .T e s t i n g - t h e - l i m i tasn d t h e
study of adult age differencesin cognitive plasticity of a mnemonic
skill. Devclopmennl Ps1'thoktg,",25, 247 256.
K y n e t t e ,D . , K e r n p e r S
, . , N o r m a n ,S . , & C h e u n g ,H . ( 1 9 9 0 ) .A d u l t s ' w o r d
r e c a l la n dw o r d r e p e t i t i o nE. x p e r i m e n t aAl g i r t gR c s e a r c h1, 6 , l 1 7 - 1 2 1 .
M o r r i s , R . G . , G i c k , M . L . , & C r a i k . F . I . M . ( 1 9 u 8 ) .P r o c e s s i nrge s o u r c e s
and age dilfbrencesin working memory. Memorv & Cognition, 16,
362-366.
Salthousc,T. A. (1980). Age and memory: Stralegiesfirr localizing the
l o s s .I n L . W . P o o n .J . L . F o z a r d ,L . S . C e r m a k ,D . A r e n b e r g &
, L. W.
Thompson (Eds.), Nex, dircctions in memort'and aging (pp. 47-65).
Hillsdale. NJ: LawrenceErlbaum Associates.
S a l t h o u s eT, . A . ( 1 9 9 0 ) . W o r k i n g m e m o r y a s a p r n c e s s i n gr e s o u r c ei n
cognitive aging.DcvclopmcntttlReviev',10. l0l-124.
S a l t h o u s cT. . A . ( 1 9 9 2 ) .W h a t d o a d u l ta g ed i f fe r e n c e si n t h e D i g i t S y m b o l
SubstitutionTest reflect'l.lournal of Gerontololit,: Ps'"r'hologitalSt'ie n c a s , 1 7P, l 2 l - P l 2 u .
S a l t h o u s eT. . A . ( 1 9 9 3 ) .S p e e da n dk n o w l e d g ea sd e t e r m i n a n tosl ' a d u l ta g e
dil'lercncesin verbal lasks. Journul ol Cerontologt': Psv-chologital
Scicnczs,1ll. P29-P36.
S a l t h o u s eT. . A . ( i n p r e s s ) .S p e e dn r e d i a t i o no l - a d u l t a g e d i t l ' e r e n c eisn
cognition. Devck4tmcntulPstcholog,t'.
S t i n e .E . L . . & W i n g f i c l d ,A . ( l 9 t l 7 ) . P r o c e s sa n d s t r a t e g yi n m e m o r yf i r r
and Aging, 2,
speechamong younger and older adults. Ps.r'cftolog1'
272-279.
T h o m p s o n ,L . A . . & K l i e g l . R . ( 1 9 9 1 ) .A d u l t a g ee f f e c t so f p l a u s i b i l i t yo n
memory: The role of time constraintsduring encoding. Journal of
Learning, Memort and Cognition, 17, 542ExperimentulPs,-r'hologl':
555.
ReceivedSeptemberI 7, I 992
AccepredApril l, 1993
Appendix
Notes
l. This particularlevel of accuracyis somewhatarbitrary, but
encodingtime estimatesbasedon diff'erentlevelsof accuracywere
highly correlatedwith one another.For example, the corelations
in the sampleof young subjectsbetweenencodingtime estimates
with accuracylevclsof l07o and 25o/owere.92 (original)and .87
(rcordcred),and .94 (original) and .94 (rcordered)betweenaccuracy levelsof 254/aand4OVa.
2. Reliabilityof the criterion memory measurecould not bc
determinedin Study I, but the estimatesin Study 2 derivedby
boosting the correlationbetweenaccuracyin the 800 and I,600
msec conditions by the Spearman-Brownformula were .83 fbr
recall in the orisinal order and .8ll for reorderedrecall.