51 (2), 102-lr2 (199s)
SwissJournal of Psvchology
Influence of ProcessingSpeedOn Adult Age Differences
in Learning
Timothy A. Salthouse
measures,might be med
working memon. In tr
signed to assessthe rper
mentary processlns opc
ministered to examine I
speedas a potential med
encesin thesetasks.
Institute
of Technology,
Atlanta,USA
Schoolof Psychology,
Georgia
Study 1
Two studies were conducted with adults from a wide range
of ages to investigate the mechanisms by which a slower
processing speed contributes to adult age differences in
short-term learning. Although statistical control of measures of perceptual speed substantially reduced the age-
related variance in measures of associative leaming and
of maze leaming, there was little evidence that speed exerted its effbcts through processesof forgetting previously correct responsesor perseveratingwith the same incorrect resDonses.
racy, and when the forgetting and speed measures were statistically controlled, the relation
between age and associative learning performance was no longer significantly different
from zero. More detailed analysesof the data
across successivetrials were reported by Salthouse & Dunlosky (in press).These analyses
identified forgetting and perseverations,which
were operationalizedas the continued use of a
response already disconfirmed for the current
stimulus, as important correlates of adult age
differences in trial-by-trial improvement.
The current project consists of two studies
designed to replicate and extend the earlier
findings. Study I examined familiar (digits and
letters) and unfamiliar (symbols) stimulus materials in associativelearning. The contrastwas
not pure becausethe familiar stimuli were presented before the unfamiliar stimuli for all research participants,but the comparison should
still prove informative.
Study 2 examined the effects of repeatedexperienceswith the sametype of stimuli (but different items) across three blocks of trials. Generalizability was investigated with another
learning task designedto be amenableto a similar type of process decomposition in terms of
forgetting of previously correct responsesand
perseverationof previously incorrect responses.
In addition, two working memory testswere administeredin Study 2 to investigatewhether the
age-related Iearning differences, either in the
Institute
onAgThisresearch
wassupported
by National
percentage correct measure or in the
overall
ing GrantR376826.I wouldlike to thankJohnDunlomore detailed forgetting and perseveration
on an earlierdraft.
sky for helpfulcomments
Adult age-related differences in learning are
well-establishedin a variety of learning situations. For example, there are reports of slower
learning by older adults in maze learning (Baddeley & Brooks, 1976;Husband,1930;Wright,
1957),minor tracking(Wright & Payne,1985),
reading degradedor inverted text (Hashtroudi,
Johnson& Chrosniak, l99l), identificationof
stimuli from fragments(Russo& Parkin, 1993),
verbal paired associates(Kausler & Puckett,
1980; Monge, 1971; Winn, Elias & Marshall,
1976), and remembering lists of unrelated
words (Crook & West. 1990: Macht & Buschke, 1983;Mueller,Rankin & Carlomusto,1979;
Query & Megran, 1983; Rankin & Firnhaber,
1986;Worden& Sherman-Brown,1983).However,despitethe robustnessofthe phenomenon,
the reasonsfor age differences in learning are
not yet understood.
A recent project (Salthouse, 1994) identified
two factors that appear to contribute to adult
age differences in associative learning; the
probability of forgening previously correct responses,and the speed with which simple operations could be executed. That is. increased
age was associatedwith greater forgetting of
responsesthat were correct on the prior trial,
and with slower performancein simple perceptual comparison tasks. Both of these measures
were correlatedwith associativelearnins accu-
t02
@ VerlasHansHuber.Bem 1995
Method
Subjects. The particrpan
formed the currenttarl.
ory tasksinvolving rerfu
tion (reported in Sahhou
170 adults complered ho
study. The subjecrsuere
paper ads, communrl\ r
tances of researchasrl\ul
rized in three age group
characteristicsare \umm
?"asks.The t\r'o pap€rspeed tasks. Letter Cor
Table l: Demographre :har+rr
pantsin Studv I
,'\
l(
fg
Age
Mean
SD
Vc Female
Years of Education
Mean
SD
Health
Mean
SD
:
Letter Comparison
Mean
SD
il:e
PatternComparison
Mean
SD
lb ,1ll
4nr
Note: Education refers to \car\
pleted, and health is a self-ran
ging from I for excellent ro 5
measures,
might be mediatedby relationswith
working memory.In both studies,tasks designedto assessthe speedof carryingout elementaryprocessingoperationswere also administeredto examinethe role of processing
speedas a potentialmediatorof the agedifferencesin thesetasks.
Study I
Method
Subjects.The participantsin this study performed the currenttasksafter a seriesof memory tasksinvolvingverbaland spatialinformation (reportedin Salthouse,
in press).A total of
170adultscompletedboth tasksin the current
study.The subjectswere recruitedfrom newspaper ads, community groups,and acquaintancesof researchassistants,
and were categorized in threeage groupswhosedemograplic
characteristics
are summarizedin Table l.
Zasts. The two paper-and-pencil
perceptual
speedtasks, Letter Comparisonand pattern
Table I: Demographic characteristicsof research participants in Study I
I 8-39
64
Age
10-59
55
60-88
5l
Age
Mean
SD
25.3
6.5
47.9
5.1
70.I
6.1
VoFemale
51.5
60.0
5l.0
Yearsof Education
Mean
SD
14.3
2.0
14.5
2.3
Health
Mean
SD
13.2
Comparison,required subjectsto decide whether pairs of letters or pairs of line patterns were
the same or different. The test forms consisted
of pairs of 3, 6, or 9 letters (Letter Comparison) or line segments (Pattem Comparison)
with a line between the members of each pau.
Subjectswere to write an S on the line between
the two members of the pair if they were the
same, and to write a D on the line if they were
different. The measures of performance were
the number of correct items minus the number
of incorrect items produced in 30 sec.
Trials in the associative learning task contained a single stimulus item on the left, and
the set of responsealternativesin a column on
the right. The subject selectedthe appropriate
responseby using the arrow keys on the keyboard to position an zurow in front of the designated response item. After the subject selected the responseand pressedENTER, feedback was presentedin the form of an auditory
tone to indicate correct or incorrect, and visual highlighting of the correct response.The responsealternativeswere then re-orderedand a
new stimulus term was presented.Three practice trials with two stimulus pairs (i.e., I-X,
II-Y) were presented,followed by two sets of
five pairs each. The first set consistedof pairs
of digits and letters (i.e., l-C, 2-A,3-D, 4-8.
and 5-B), and the second set consistedof pairs
of unfamiliar symbols. (SeeSalthouse,1994 for
an illustration of the symbol stimuli). Trials
continued until the subject reached a criterion
of three trials with all pairs correct, or until a
maximum of l0 trials had been presented.
Resuhs
L.)
) o
) 1
0.9
1.0
) .1.
i:0
LetterComparison
Mean
SD
I 1.59
3.02
9.53
3.21
? ir
3:o;
PatternComparison
Mean
SD
r8 . 3 0
4.04
15.69
3.40
t2.33
a.z I
Note: Education refers to years of formal education completed, and health is a self-rating on a 5-point scale ranging from I for excellent to 5 for poor.
Percentagecorrect across successivetrials for
the three age groups is illustrated in Figure l.
Notice that all age groups begin at rhe same
level, but that the amount of improvement is
inverselyrelatedto age.It can also be seenthat
the pattern was very similar with both sets of
stimuli.
An Age (18-39, 40-59, 60-88) x Stimuti
(Familiar,Unfamiliar)x Trial (1 through lO;
ANOVA revealedsignificant(i.e.,p <.01) effectsof Age, F(2,167)= 22.24,MSe = 0.67;
Trial, F(9,1503) = 114.04, Age x Trial,
103
F ( 1 8 , 1 5 0 3 )= 4 . 9 1 , M S e = . 0 4 5 ,a n d S t i m u l i x
Trial, F(9,1503) = 2,79, MSe = .037. The laG
ter interaction reflected slightly greater improvement with familiar stimuli. Becausemany
subjectsreached the criterion in fewer than 10
trials and did not continue in the task, values
of lo07o were usedto replacetheir missing values in this initial analysis. A similar analysis
was conducted with only data from trials I to
6 in which no missing observationshad to be
replaced.The same general pattern was apparent in this analysisas there were significant eff e c t s o f A g e , F ( 2 , 1 6 7 ) = 1 7 . 7 3 ,M S e = . 3 1 ;
Trial, F(5,835) = 82.00, and Age x Trial,
F(10,835)= 3.93, MSe = .046. Only the Stimulus x Trial interaction was not significant in
the analysisrestrictedto trials I through 6, suggesting that the advantage of familiar stimuli
occurred primarily in later trials.
The pattern in Figure l, and the significant
Age x Trials interaction, indicate that there
were significant age differences in the rate of
learning. The learning data were therefore examined with path analysis techniquesto determine where distinct age-relatedeffects occur.
Figure 2 illustratesthe path model usedto guide
the analyses.Notice that performance on each
trial was assumed to be affected by performance in the immediately preceding trial, and
possibly also by age.
Age-related effects were not expectedon the
first trial becausethere was no prior opportunity to learn since it was the initial exposureto
the materials fbr all subjects. The interesting
question was where in the sequencedo unique
age-relatedeffects occur. If age-relatedeffects
occur only on early trials, then all of the agerelated influences on later trials can be presumed to be mediated throush the effects on
Digit - Letter
Table 2: Path coefficienr! tor
cessive trials, Study I
Path
possible
from
Figure2: Pathdiagram
illustrating
relations
ageto performance
on successive
trials.
the first few trials. However, if there are independent age-relatedeffects on later trials, then
distinct age-relatedinfluences can be inferred
to be operating on processesat several different stagesof learning.
The standardizedpath coefficientsfrom these
analysesare presentedin Table 2. As expected,
there were strong relations between the measures of performance on successivetrials, particularly after the first trial (lower half of table). Significant age effects were found on trials
2 and 4 with familiar stimuli, and on trials 2,
3, 5, and 6 with unfamiliar stimuli. The pattern
is somewhatvariable,but it is clear that the agerelated influences are not restricted to the first
one or two trials on the task. Instead,there appear to be independent age-related effects at
s e v e r a lp h a s e si n l e a r n i n g .
The processesinvolved in the age effects in
learning were next investigated by computing
measuresof forgetting and perseveration.The
former is the number of forgets (correct on trial
n followed by incorrect on trial n+1 for a given stimulus) divided by the total number of correct responseson the previous trial, and the latter is the number of perseverations(the same
incorrect responseto a given stimulus on two
Symbol - Symbol
1@
s
o
o
s
o
m
o
&
6
o
I
o
(L
Trial
104
Figure ): Percentagecorrect in
associative learning with
familiar and unfamiliar stimuli
across successivetrials in three
age groups, Study L
a
b
c
d
e
f
Age -l
Age -2
Age -3
Age -4
Age -5
Age -6
h
2-3
|
5-4
j
k
4*s
5*6
D r g r t sL- . ' t
1 t
08
-:l'
-.l6
-.08
.19
..19'
.-i6,
.5li'
* Different from zero br m.'rr
successivetrials) divided
of errors on the previous
each case the relevant r a
percentageof the possible
type of response.
The overall meansfor rl
were 38.4Vofor the tamih
for the unfamiliarsrimulr.
severationmeasure\ \\ ere
iar stimuliand l-5.1? tt'r
uli. The same type of Ar
ANOVA conducted on th
measurewas conductedc
perseverationmeasurei el
only 5 trials becauseneirh
severationcan occur untrl
prior trial. With the forsen
F ( 2 , 1 6 7 )= 5 . 3 0 .\ l S e = l
= 3 . 6 3 ,M S e = . 1 - 5 .e t i - e c
no other main effect. or I
nilicant. With the perse\c
t h e A g e , F ( 2 . 1 6 1 t= - i . f > l .
uli, F(1,167=
) l - 1 . 8 1\.l S
significant. Both foreerrrn
were more frequent urth r
ting decreasedacrossrnal
occurred more often u rth
with unfamiliar stimulr
Becausethe onll srgnr
with the forgetting measun
eraged across the first srr
to improve reliabilitr. Tt
Letter Comparison and
paper-and-pencilspeed rn
thus a composite speed rne
averagingthe two z-scorfi
Table 2: Path coefficientsfor influencesof ase on successivetrials,Study I
Path
a
b
c
d
e
f
Age -l
Age -2
Age -3
Age -4
Age -5
Age -6
o
l-)
h
2-3
i 3 4
4-5
J
k
5-6
Digits-Letters
Symbols-Symbols
-.14
-.24*
-.08
-.21*
-.16
-.08
-.07
-.24*
-.29*
-.04
-.lg*
_.26*
.19
.39x
.41*
.56*
.58*
.27*
.49*
.58*
.60*
. 5 5*
Figure3: Pathdiagram
possible
illustrating
relations
from
andmeasures
of associative
learningperforl%:"..r0".0
* Different from zero by more than 2 standarderrors.
successivetrials) divided by the total number
of errors on the previous trial. Notice that in
each case the relevant value is expressedas a
percentageof the possibleopportunitiesfor that
type of response.
The overall meansfor the forgetting measure
were 38.4Vofor the familiar stimuli and 36.6Vo
for the unfamiliar stimuli, and those for the perseverationmeasureswere lJ.67a for the familiar stimuli and 15.2Vofor the unfamiliar stimuli. The same type of Age x Stimuli x Trial
ANOVA conducted on the percentagecorrect
measure was conducted on the forgetting and
perseverationmeasuresexcept that there were
only 5 trials becauseneither forgetting nor perseverationcan occur until there was at leastone
prior trial. With the forgetting measure,the Age
F(2,167)= 5.30, MSe = .28: and Trial, F(5,835)
= 3.63, MSe = .15, effectswere significantbut
no other main effects or interactions were signiiicant. With the perseverationmeasure,only
the Age, F(2,16'l)= 5.64, MSe = .10, and Stimuli, F(1,167) = 13.82,MSe = .07, effectswere
significant. Both forgetting and perseverations
were more frequent with increased age, forgetting decreasedacrosstrials, and perseverations
occurred more often with familiar stimuli than
with unfamiliar stimuli.
Becausethe only significant trial effect was
with the forgetting measure,the values were averaged across the first six trials in an attempt
to improve reliability. The correlation of the
Letter Comparison and Pattern Comparison
paper-and-pencilspeed measureswas .57, and
thus a composite speedmeasurewas createdby
averagingthe two z-scores.Interrelationsofthe
measureswere then examined in path analyses
with the path model illustrated in Figure 3.
Path coefficients for both sets of stimuli for
the model illustrated in Figure 3 are presented
in Table 3. The following points should be
noted about the results in this table. First, as
expected, there were strong negative relations
between age and speed. Second, age and percentage conect were negatively related, although with both sets of stimuli the direct effects were smaller than the total age-relatedeffects (i.e., the path coefficients were -.14 and
-.21 compared to the correlations of -.34 and
-.42, respectively).The reduced age relations
Table3: Pathcoefficients
for interrelations
of age,perceptual speed, percentage forgetting, percentage perseverations, and percentagecorrect, Study I
Path
a
b
c
d
Age Age Age Age -
Speed
PercentageCorrect
Forget
Perseveration
e Speed- PercentageCorrect
f Speed- Forget
g Speed- Perseveration
h Forget- PercentageCorrect
- Percentage
i Perseveration
Correct
Correlations
Age - Percentage
Correct
Age * Forget
Age - Perseveration
Forget- PercentageCorrect
Perseveration- Percentage
Correct
DigitsLetters
SymbolsSvmbols
-.59,k
-.14*
.04
.l 8
-.59*
-.21x
.14
.03
.07
.06
la
1?*
-.06
*.09
-.41*
-.54*
-.51*
-.36*
-.34*
.l I
.21*
-.45*
-.59*
-.42+
.30*
.08
-.65*
,.47*
* Different from zero by more than 2 standarderrors.
105
Table4: Demographicand performancecharacteristicsof
researchparticipantsin Study 2
2o-39
38
Age
4o-59
42
60-79
Age
Mean
SD
30.8
6.0
48.0
5.8
68.9
51t
VoFemale
68.4
42,9
+r'z
Yearsof Education
Mean
SD
1 5I.
1.5
l6.l
t:
Health
Mean
SD
1.7
0.7
5t
2.4
I
2.3
1.2
? l^
l'0
PatternComparison
Mean
SD
t7.'tI
3.64
r6.60
2.67
tr.,|?
Letter Comparison
Mean
SD
10.62
3.01
9.89
1.43
6^19
2'90
Digit Digit RT
Mean
SD
624
t04
689
t49
Digit SymbolRT
Mean
SD
1243
270
t425
318
ii9
,?11
ReadingSpan
Mean
SD
2.68
|.28
2.92
1.22
I ]q
l'00
ComputationSpan
Mean
SD
4.42
1.63
4.64
2.tl
3.34
1.48
for that task. Each task was administeredtwice
with different stimulus items to increase reliability.
The two computer-administered reaction
time tasks each involved the presentationof a
pair of items in the middle of the screen and a
code table with nine pairs of items at the top
of the screen.In the Digit Digit task the code
table containedidentical pairs of digits, and the
decision concerned whether the two digits in
the middle of the screen were physically identical. In the Digit Symbol task the code rable
contained pairs of digits and symbols, and the
decision concerned whether the digit-symbol
pair in the middle of the screen matched according to the code table. Ninety trials were
presented
in eachof two administrations
of the
tasks, and becauseaccuracyaveragedover
95Vo,medianreactiontime in millisecondswas
used as the measureof reaction time oerformance.
The associativelearning task was very similar to the version with symbols used in Study
I except that three sets of six trials each were
administered, with six different stimulus pairs
in each set. The maze leaming task also involved six trials on each of three different
mazes.In the computer-administeredmaze task
the subjects used the arrow keys on the keyboard to move a cursor through a grid in which
obstacles(i.e., the 7o symbol) blocked many of
the possible paths. Only a 3x3 portion of the
10x10 grid was visible at any given time, but
the goal was to find the end point (designated
by the digit l) in the minimum number of
moves. Each maze contained six choice points
with three options each (e.g., if moving down,
the options were left, right, and straight or
down). The associative learning and maze
learning trials were presented in alternating
blocks, with six trials of one set of associative
learningstimuli, six trials of one maze,six rrials
of a second set of associativelearning stimuli,
six trials of a secondmaze,and finally, six trials
of the third set of associativelearning stimuli
and six trials of the third maze.
Because the sessions were limited to 2.5
hours in length, some researchparticipants did
not complete all of the tasks. Furthermore, the
attrition was not random becausethe fastestand
most efficient subjectswere more likely to finish every task. These were disproportionately
the young subjects because 35 of 38 young
adults completed all three setsin the maze task,
but only 16 of 37 older adults completed all
three maze sets. Only results from subjects
completing all three sets will be emphasizedin
the discussion,but results are reported from the
subjectscontributing data to each set.
Results
Two possible measures of performance were
available in the maze task, the number of keystrokesand the percentageof correct choices at
the three-alternative choice points. The two
107
measureshad moderatenegative correlations
with one another,as the numberof keystrokes
decreased
from 122on trial 1 to 51 on trial 6,
and the percentagecorrect values increased
from 327oon trial I to 67Voon trial 6. The median correlationbetween the two measures
acrosstrials was-.47, with a rangefrom -.27
to -.63. Neithermeasureis optimalbecause
the
number of keystrokes measure is highly
skewed,andthereareonly six choicepointsfor
the percentage
correctmeasure.However,becausethe percentage
correctmeasureis more
comparableto the measurein the associative
learningtask, and is amenableto decompositions into forgettingand perseverations,
it was
usedin subsequent
analyses.
Percentage
correctby trial and agegroup for
the threesetsof trials in the associative
learning and mazelearningtasksare illustratedin
Figure4. Noticethat the patternwith the associativelearningdatais very similar acrossthe
threesuccessive
setsof stimuli,and resembles
that observedin Study l. The datain the maze
learningtaskswere more variable,but trial-bytrial improvementand age differencesare still
apparent.
An Age (18-39,40-59, 60-79) x Set (first,
second,third) x Trial (l through6) ANOVA
was conductedwith eachlearningtask for the
subjectswith completedata.In the associative
learningtaskthereweresignificantmain effects
of Age, F(2,86) = 6.00, MSe = 0.41; Set,
F(2,172) = 63.31, MSe = .09; and Trial,
F(5,415)= 82.76,and a significantinteraction
of Age x Trial, F(10,830)= 3.19,MSe = .04,
with Set.In the mazelearnbut no interactions
ing task there were significant main effects of
Set, F(2,1661= 40.36, MSe = .06; Trial,
F(5,415)= 92.80,MSe = .03;anda significant
F(10,830)= 2.35,MSe
Setx Trial interaction,
= .02,but no interactions
with age.Neitherthe
maineffectof Age,F(2,83)= 2.17,nor theAge
x Trial interaction,F(10,830)=2.09), weresignificantin the mazelearningdata.
SeparateAge x Trial ANOVAs were also
conductedon the datafrom eachset to exam-
Maze
PairedAssociates
^ - gl4287
70
&
50
ao
s
20
ine patterns with the gre:
of subjectsin each compa
interactions were signific
except the Age x Trial rn
1.44,for the first maze.T
indicate that, as expecred
evident in the measure
tasks.
Path analyseswere ne\
the path diagram illusrra
path coefficients are pr
where it can be seen thal
ilar to that in Studl l. Thi
erate to large relations trt
rect on successivetrials.a
age effects on later tnals
er, these latter results sue
tinct age-relatedinfl uencc
independent of the eft-ecr
Age x Set x Tnals A\r
ducted on the percenrae
centage perseveratronrnc
getting measurein the ass
the following effecrs rr
F ( 2 , 1 6 6 )= 1 9 . 3 . 1N. { S e=
8 . 4 9 ,M S e = . 1 3 . b u r n o r
M S e = . 2 1 ,n o r a n v i n r e r
Forgetting percentagesd
c e s s i v es e t s ( i . e . . - l : . 8 . .
acrosssuccessivetnals l
30.5, and 26.8). r*'irh tk
perseveration measure tl
w e r e A g e , F ( 2 . 8 6 r= 6 . 1 1
x Set, F(4,166) = -3.66.!
o
Table5. Parh..*rTr
C)
q)
o)
G
o
o
(L
s
Path
20
10
1m
^ - 351371t7
m
&
70
60
$
40
s
10
3
4
Trial
108
5
6
Figure 4: Percentagecorrect ln
associative
learningand maze
leamingacrossthreesuccessive
trials in three age groups,Study
2. Successive
setsof stimuliare
representedin the panelsfrom
top to bottom.
a
b
c
d
e
f
A g e- l
Age -l
Age --1
Age{
Age -5
Age -6
g
h
i
I-l
2-3
3-4
i
4-S
k
5-6
*p<.ol
ine patternswith the greatestpossiblenumber tions were more frequent with increased age
of subjectsin eachcomparison.
All Age x Trial (i.e., 10.6, 12.6, and 17.6), but this trend was
interactionswere significantin theseanalyses less pronounced in the third set. The only sigexceptthe Age x Trial inreractionF(10,570)= nificant effects on the forgetting measurein the
1.44,for the first maze.Theseresultstherefore maze learning task were Set, F(2,166) = 25.67,
indicatethat,asexpected,agedifferenceswere MSe = .08; and Trial, F(5,415)= 4.07, MSe =
evident in the measuresof learning in both .05; but not age (F < 1) or any interactions.
tasks.
These effects were attributable to lower levels
Pathanalyseswerenext conductedbasedon of forgettingin set 2 (i.e., 19.7,8.1, 19.6),and
the path diagramillustratedin Figure 2. The a slight decreasein forgetting percentageacross
path coefficientsare presentedin Table 5, successivetrials (i.e., 19.0,18.0,14.9,12.4.and
whereit can be seenthat the patternwas sim- 14.8).No significant effects were evident in the
ilar to that in Study 1. That is, thereweremod- maze leaming perseverationmeasures.
erateto largerelationsbetweenpercentage
corCorrelations between similar speed and
recton successive
trials,andseveralsignificant memory measures were .63 between the two
age effectson later trials.As mentionedearli- paper-and-pencilspeed measures,.69 between
er, theselatterresultssuggestthattherearedis- the two reaction time speedmeasures,and .39
tinct age-related
influenceson learningthat are between the two working memory measures.
independent
of the effectson early trials.
The correlation for the working memory meaAge x Set x TrialsANOVAs were alsocon- sures was not very high, but it increasedto .53
ducted on the percentageforgetting and per- after adjusting for the reliability of the meacentageperseveration
measures.
With the for- sures.Compositemeasureswere formed for use
gettingmeasurein the associative
learningtask in subsequentanalyses by averaging z-scores
the following effects were significant: Set, for the relevant measures.
F(2,166)= 19.34,MSe = .01;Trial,F(5,415)=
The age-relatedvariance in the averageper8.49,MSe = .13,but norAge,F(2,86)= 3.34, centagecorrect measuresacrossall 6 trials was
MSe = .21,nor any interactions
involvingage. computed before and after control of the speed
Forgettingpercentages
decreasedacrosssuc- and working memory composite measures.Recessivesets(i.e.,42.8, 31.3,and 28.1),and sults of these analysesare presentedin Table 6,
acrosssuccessive
trials(i.e.,45.9,35.1,31.9, where it can be seenthat there was large atten30.5, and 26.8).With the associativelearning uation of the age-relatedeffects after statistical
perseveration
measurethe significanteffects control of the speed measures.For example,
wereAge,F(2,86)= 6.12,MSe = .07,andAge control of the perceptual speed composite rex Set,F(4,166)= 3.66,MSe = .04.Persevera- duced the age-relatedvariance in set I accuraTable5: Pathcoefficientsfor influencesof aqeon successive
trials.Studv2
t
(n)
Age -l
Age -2
Age -3
Age -4
Age -5
Age -6
a
b
c
d
f
o
a
h
i
j
k
.
3
l-')
-
2-3
4
4-s
5-6
Paired Associates
Set
)
?
(lr7)
-.t5
-.30x
-. l8*
-.21x
*.32*
-.24*
(lll)
-.15
-.32*
-.24*
-.04
-. l6*
-.15*
-.03
-.24*
-.f8*
-.07
-.l l
-.10
.07
.35*
.38*
.30*
.49*
.20*
.59*
.69*
.68x
.69*
.t2
.61*
.64*
.64*
.73*
(89)
Maze Learning
Set
A
i
(89)
-.04
-.33x
-.14
-.03
-.10
- . 1l *
.30x
.69x
.78*
.84'k
.84*
l
I
(rr7)
2
3
A
(86)
(e8)
-.05
-.20*
.02
-. 13
-.07
-.21*
.15
-. l6
-.29*
.08
-.28*
-.01
.21*
.51*
. 51 *
.6f*
.50i
.19
.45*
.71*
.54*
.'ll*
-.05
-.14
-.16
*.21*
-.10
-. l0
.16
.36*
.60*
.68*
.69*
i
l
(86)
.0s
-.t9
-.15
-.04
-.0'7
-.05
.24*
.65*
.7'7*
.78*
.83x
*p<.ol
r09
Table6: Proportionsof age-relatedvariancein averagepercentagecorrectbeforeand after control of speedand working memory, Study 2
Set
Age Alone
PSpd
RTSpd
After Control of:
WMem
PSpd
WMem
I
2
3
All
(n=117)
(n=ll1)
(n=89)
(n=89)
.238*
.204*
. 10 7 *
.152*
AssociativeLearning
.049* .095*
.153*
.043
.081* .169*
.021
.024
.068
.035
.064
.l 11*
I
2
3
All
(n=ll7)
(n=98)
(n=86)
(n=86)
.058*
.070*
.ll0*
.061
Maze Leaming
.000
.008
.017
.010
.003
.019
.055
.075*
.069
.01l
.02t
.030
RTSpd
WMem
PSpd
RTSpd
WMem
.014
.029
.076*
.078*
.019
.057
.038
.038
.006
.024
.000
.003
.042
.005
.002
.001
.064
.015
.000
.001
.051
.007
.037
.u2
*p<.ol
Note: PSpd refers to the perceptualspeed(Letter Comparison,Pattem Comparison)composite, RTSpdrefersto the reactiontime speed(Digit Digit, Digit Symbol)composite,andWMem
refersto the working memory(ReadingSpan,ListeningSpan)composite.
cy by 79.4Vofor associative learning, and by
l00%ofor maze learning. It is also apparentin
Table 6 that the reduction ofthe age-relatedvariance was only slightly greaterwhen both speed
and working memory were controlled than
when only the speedmeasureswere controlled.
The age-related variance in the composite
working memory measure was also examined
after control of the composite speedmeasures.
The R2 associatedwith age was reduced from
.141 to .031 (78.0Eo)after control of the perceptual speed composite, and to .029 (79.4Va)
after control of the reaction time speed composite. The initial values were clearly significantly greater than zero, but neither of the residual values differed significantly from zero.
Furthermore, when the order of the variables
was reversed, there was a much smaller percentagereduction of the age-relatedvariance in
the composite speed measure after control of
the composite working memory measure.
These results are consistent with the results of
several earlier studies (e.g., Salthouse& Babcock, l99l; Salthouse,l99l1'1992), and help
explain why there is little additional reduction
of the age-related variance in the learning
measures when working memory was controlled after speed.That is, most of the age-related variance in the working memory measures appears to be shared with the speed measures.
110
Path analyseswere next conductedto examine interrelations among the variables accoroing to the path model in Figure 5. Coefficients
with the perceptual speed composite are presentedin Table 7. The pattern with the RT speed
composite was very similar, and thus only results with the perceptualspeedmeasureare presented.
The following points should be noted about
the data in Table 7. First, the relations of age
to the measuresof speed,working memory, and
percentage correct are weaker in successive
blocks, which may partly be a consequenceof
selective attrition in which progressively more
lower-ability older adults fail to continue in
successivesets. Second, as in Study l, there
were moderate to large paths between age and
Table 7: Path coettl.-r
forgetting, percenrarc
Path
a
b
c
d
e
Age Age Age Age Age -
f
g
h
i
SpeedSpeedSpeedSpeed-
j
k
I
W M e m .- P e r ( WN{em. Ftrrrc
WMem. - Per.cr
m
n
Forget- PcrC,r
P e r s e r-. P e r C , v
Speed
tA'\lem
Per.Ct'r
Forger
Perrer
\\'\terr.
PerCrv
Forscr
Per.er
Correlations
Age - Per.Cor
Age - W\tem
Age - Forger
Age - Perser
*p<.ol
speed (a), betueen percr
percentagecorTectI m r. a
age perseveratlon\ and F
However, there ua: lrttle
between age and Frer
between age and percent
between age and percenu
(e). This pattern \usseil!
effects may' be medrarcd
and working memon rnfl
ting and perseverattonm
l. however. t'eu parh. i
working memory ro thc pr
getting, or perse\erarl(rn
cantly different from 1er'
General Discussion
Figure5: Pathdiagramillustratingpossiblerelationsfrom
age to speed,working memory,and measuresof associative leaming performance.
The results of these irudt
tence of adult age differer
with different kinds of .
learning, and in a new ma
thermore, the age-related
simply restrictedto the i
Table 7: Path coefficientsfor interrelationsof age, perceptualspeed,working
- memory, percentage
forgetting,percentageperseverations,
and percentagecorrect,Study 2
Paired Associates
Set
Path
t
(nl
z
-
f
MaTe Learning
Set
Ail
(8e)
I
(rl7)
2
(98)
3 A i l
(86) (86)
(l17)
(lll)
-.65*
_.23x
.10
-.05
-.68*
-.13
-.1I
.12
. 3 1*
.22*
.18
-.09
-.35x
.31x .20
.20
.15
.06
.06
-.06 -.0'7 -.07
-.t7
- . 3 7 * -.49*
.22* .28,k
.02
.09
-.30* -.06
-.22 -.00
.14* .t2
. 0 5 .l 0
-.03 -. l8 , . 1 0 - . 0 8
-.r3 -.18 -)6* _))*
(8e)
a
b
c
d
e
Age Age Age Age Age -
f
g
h
i
SpeedSpeedSpeedSpeed-
j
k
I
WMem.- Per.Cor. .13
WMem. - Forget
.10
WMem. * Persev. -.12
-.05
-.1'7
.14
m
n
Forget- Per.Cor.
Persev.- Per.Cor.
-.21*
-.21*
-.53x -.59* -.58*
-.3-5* -.40* -.3'7*
-.5I *
* . 5 1*
-.49*
-.39*
.t2
.22
-.45* -.33* -.39x
-.35x ,.26 -.26
.23
.30* . 3 3 *
.38* .lI
.35*
-.24* -.26* -.33*
-.38* -.36* -.26
.23
.t'7 . t 4
.16
. 2 2 .30x
Speed
WMem.
Per.Cor.
Forget
Persev.
WMem.
Per.Cor.
Forget
Persev.
Correlatktns
Age - Per.Cor
Age - WMem
Age - Forget
Age - Persev.
_ )5*
</*
-.54*
-.15 -.15
* . 0 8 -.05
.20
-z-)
-.12
.0'l
-.01
-.26*
-.12
-.02
.\a*
-.08
<2*
<l*
-.65* -.65x
-.23* -.18 - . 1 5 - . 1 5
.02
.03 -.07
.02
.02
.01
.12 .0'7
-.03
.1 5
.23 .12
.2t
.21
.08
.05
.03 -.t'7
.0t -.13
-.53* -.50* -.47*
- . 4 7 * ,.41* -.59*
-.26
-.26
.18
.25
*p<.ol
speed (a), between percentage forgetting and
percentagecorrect (m), and between percentage perseverationsand percentagecorrect (n).
However, there was little or no direct relation
between age and percentage correct (c),
between age and percentageforgetting (d), or
between age and percentageof perseverations
(e). This pattern suggeststhat rhe age-relared
effects may be mediated through small speed
and working memory influences on the fbrgetting and perseverationmeasures.As in Study
l, however, few paths fiom either speed or
working memory to the percentagecorrect, forgetting, or perseverationmeasuresare significantly different from zero.
General Discussion
The results of these studies confirm the existence of adult age differences in learning, both
with different kinds of stimuli in associative
learning, and in a new maze learning task. Furthermore, the age-relatedinfluences were not
simply restricted to the initial learning trials,
and thus they represent genuine effects on
learning rather than merely effects on processes involved in the first one or two exposuresto
the stimuli. Speed,and to a lesserextent, working memory, appearsto be involved in the age
difference in learning because the age-related
variance in the percentage correct measures
was considerably reduced after statistical control of measuresof speedand working memory.
Findings by Salthouse& Dunlosky (in press)
were also replicatedin the discovery that learning is poor when the subjectsfail to retain feedback as reflected by forgetting (i.e., not repeating correct responses),and perseverations(i.e.,
continuing to repeatincorrect responses).Howeveq it is not simply the case that the age effects in learning occur becauseof a greaterproportion of forgetting and perseveration responsesbecausethe relations between age and
these measures were relatively small. Moreover, the weak and inconsistent relations
between the speed measuresand the presumably more detailed measures of associative
learning performance indicate that the mechalll
nismsresponsible
for the speedinfluenceshave
not yet beenidentified.
In summary,adult age differencesin learning have been confirmedby the existenceof
significantinfluencesof age on measuresof
performancefrom later trials after performance
on the immediatelyprecedingtrial were taken
into consideration.Furthermore,analysesrevealedthat the speedwith which simplecomparisonoperations
couldbe executed,
theprobability of forgetting previously correct responses,
andtheprobabilityofperseverating
on
previouslyincorrectresponses,
all contributed
to trial-by-trial improvementin these tasks.
What is not yet clear,and which shouldbe a
goal for future research,is exactlyhow these
measures
are interrelated
with one anotherand
with increasedagein adulthood.
Query,W.T. & Megran,J. (1983).Age-relatednormsfor
AVLT in a male patient population.Journal of Clinical Psychology,,19,
136-138.
Rankin,J.L. & Firnhaber,S. (1986).Adult agedifferences
in memory:Effectsof distinctiveandcommonencodings.ExperimentalAging Research,12, 141-146.
Rushton,J.P.,Brainerd,C.J. & Pressley,M. (1983).Behavioraldevelopment
andconstructvalidity:The principle of aggregation. PsychologicaLBuLletin, 94,
l 8-38.
Russo,R. & Parkin,A.J. (1993).Age differencesin implicit memory:More apparentthan real. Memorv &
Cognition,2l,'73-80.
Salthouse,
T.A. ( 1991). Mediationof adultagedifferences
in cognitionby reductionsin working memory and
speed of processing.Psychoktgical Science, 2,
l 79-t 83.
Salthouse,
T.A. (1992).Influenceof processingspeedon
adult age differences in working memory. Acta
PsychoLogica,79, 155-170.
Salthouse,T.A. (1994).Aging associations:
Influenceof
speedon adultagedifferencesin associative
leaming.
Journal of ExperimentalPsychology:Learning, Memory and Cognition,20,1486-1503.
T.A. (in press).Differentialage-related
Salthouse,
influenceson memoryfor verbal-symbolic
informationand
visual-spatialinformation.Journal of Gerontology'
References
P sychological Sciences,
Salthouse,
T.A. & Babcock,R.L. (1991).Decomposing
Brooks,D.N. & Baddeley,
A.D. (1976).What canamneadult age differencesin working memory.DeveLopsic patientsleam?Neuropsychologia,
14, lll-122.
mental Psychologv,27, 763-'7'16.
Crook, T.H. & West,R.L. (1990). Name recall perfor- Salthouse,
T.A. & Coon,VE. ( 1994).lnterpretation
of difmanceacrossthe adult lif'e-span.British JournaLo.f
ferentialdeficits:The caseof agingand mentalarithPsychology,I 1, 335-349.
metic.Journal of ExperimentalPsychoktgy:Learning,
Hastroudi,
S., Chrosniak,
L.D. & Schwartz,
B.L. (1991).
Memory,and Cognitktn,20, ll'72-1182.
Effectsof agingon primingandskill learning.Psychol- Salthouse,T.A. & Dunlosky,J. (in press).Analysesof
ogy and Aging,6,605-615.
adult agedifferencesin associativeleaming.Zeitschrift
Husband,R.W. (1930).Certainage effectson mazeperJiir Psychobgie.
formance. Journal ot Genetic Psychology, 37,
Winn,F.J.,Elias,J.W.& Marshall,PH. (1976).Meaning125*328.
fulnessand interference
as factorsin paired-associate
judgKausler,D.H. & Puckett,J.M. (1980).Frequency
learning with the aged,.Educational Gerontologl,,l,
mentsand correlatedcognitiveabilitiesin young and
297*306.
elderlyadults.Joumalof Gerontology,35,3'16-382.
Worden,P.E.& Sherman-Brown,
S. (1983).A word-freMacht, M.L. & Buschke,H. (1983).Age differencesin
quencycohort effect in young versuselderly adults'
cognitive effort in recall. Journal of Cerontoktg-r-,
38,
memory for words. DevelopmentalPsychology,19,
695-700.
52l -s30.
Monge,R.H. (1971).Studiesof verballearningfrom the Wright, J.M. von (1957).An experimentalstudy of hucollegeyearsthroughmiddle age.Journal of Gerutnman serial learning.SocietasscientiarumFennica,
tology, 26, 324-329.
humanarumlitterarum,23, No. l.
Commentations
Mueller,J.H., Rankin,J.L. & Carlomusto,
M. (1979). Wright,B.M. & Payne,R.B. (1985).Effectsof agingon
Adult agedifferences
in freerecallasa functionof basex differencesin psychomotorreminiscenceand
sis of organizationand methodof presentation.
Jozrtracking proficiency.Journal of Gerontologv,40,
nal of Gerontology,
34,375-380.
t79-184.
Address oJ correspondence :
Timothy A. Salthouse, School of Psychology, Georgia Institute of Technologl,, USA-Atlanta, GA 30332
112