J ournal of Gerontolo$t : P SYCHO LOGICAL
1996. Vol. 5lB. No. l. P30-P42
SCI ENC ES
CoprriEht 1996 b' The Cerontological Socieh*of America
Generaland SpecificSpeedMediation
of Adult Age Differencesin Memory
Timothy A. Salthouse
School of Psychology, Georgia Institute of Technology.
The proposal that age-related differences in some measures of speed of performance ma! not be independent of the
age-related differences in other measures of speed of performance has been associated with considerable controiersy.
Because converging evi.dence can often resolve this tlpe of controversy, correlation-based procedures are proposed-to
distinguish general(or common) and speciftc (or unique) age-related influences on measures ofspeeded pirfurmance.
Resuksfrom three earlier studies andfrom a new stud! suggest thal a Inrge proportian ofthe age-related vapiance in a
wide range of speed measures is shared and is not distinct. Furthermore, the common or general speedfactor appeors
to play an important role in the mediation of age-related dffirences in memory.
relationsbetweenage and measuresof epiNEGATIVE
-r-\ sodic memoryperformanceare well documented(e.g.,
Craik & Jennings, 1992;Hultsch & Dixon, 1990;Kausler,
1994; Salthouse,l99l), as are negativerelationsbetween
age and various measures of processing speed (e.g.,
Salthouse,1985;Cerella, 1990). However, recentresearch
has revealed that the two sets of age relations are not
independentbecausestatisticalcontrol of the presumably
simpler speedmeasuresleadsto a largereductionin the agerelated variance in measuresof memory.
This phenomenoncan be illustratedwith datafrom a study
recentlyreportedby Salthouse(1993a)in which 305 adults
from a wide rangeof agesattemptedto remembertwo lists of
l2 words each, and also performed two speededpaper-andpencil perceptualcomparisontasksinvolving pairsof letters
or pairs of line patterns. All measureswere converted to zscores, and then the two paper-and-pencilperceptual speed
measures were averaged to form a composite perceptual
speed measure, and the recall accuracy scores on the two
lists were averagedto form a compositememory measure.
The age-related variance in the recall memory score was
.162, but it was reduced nearly 84Vo, to.026, after the
perceptualcomparison speedmeasurewas controlled.
This same pattern of attenuatedage-relatedeffects after
control of speed has been observed in a variety of other
memory measuresin a number of studies in my laboratory
(e.g., Salthouse,1994b, 1994c, 1995; Salthouse& Babcock, l99l; Salthouse,Fristoe,& Rhee, in press),and in
studiesfrom otherresearchers
(e.g., Bors & Forrin, in press;
Hultsch, Hertzog, & Dixon, 1990; Park et al., 1994). Furthermore, the results have been consistent across
experimenter-pacedand subject-pacedmemory tasks, across
tasks involving verbal and spatial information and across
different typesofexperimentalprocedures(e.g., free recall,
pairedassociates).
Because the perceptual comparison tasks used in these
studies are very simple, I have hypothesizedthat the speed
influence occurs becausethe paper-and-pencil speed measures index a fairly general processing speed factor that
decreaseswith increased age. The perceptual speed meaP30
sures are therefore assumedto reflect the speedwith which
many operations,including thosedirectly relevantto memory functioning, can be carried out. An implication of this
assumptionis that a large amount of the age-relatedvariance
across many different speed measures will be shared. A
major purpose of the current study is to investigate this
implication by examining the fit of a structural equation
model with a single latent speedconstructrelatedto all of the
individual speedmeasuresand with relationsfrom ageto the
latent speedconstruct and to each individual speedmeasure
(cf., Kliegl & Mayr, 1992; McArdle & Prescott, 1992;
Salthouse, 1994a).The hypothesisof a general speedfactor
would be supportedif there is a good fit of the datato a model
in which the latent speedconstructhas strong relations to the
individual speedmeasures,but those measureshave only
weak direct relations from age.
The speed measuresexamined in the current study were
designed to involve different types of information (i.e.,
alphanumeric,verbal, and spatial),different modesof stimulus presentation(i.e., computer-administeredvisual displays, printed, and auditory), and different modes of response(i.e., manual reactiontime, written, and vocal). In
addition to measureshypothesizedto representgeneral processing speed, measuresdesignedto reflect the speed of
specific memory-relevant processes such as search, rehearsal, association,reorganization, and retrieval were also
included. If those measuresare more directly related to the
functioning of memory than other speed measures, they
might be expectedto have strong relations to the measuresof
memory, but possibly weak relations to the hypothesized
general factor.
Two analyseswere also conductedto examine the degree
of independenceof age-relatedinfluences on measureshypothesizedto reflect the speedof specific processes.A total
of nine derived measuresthat could be postulated to represent the time for specific cognitive operations were investigated. For example, a measure of substitution time was
derived from the difference between the average of lettersymbol and symbol-letter reaction time and the average of
letter-letter and symbol-symbol reaction time. That is, in the
AGE, SPEED.ANDMEMORY
first two tasks the choice decision required determining the
correspondencebetweenletters and symbols, whereasin the
latter two tasksthe decisionwas basedon physical identity,
and hencethe difference betweenthe two variables could be
interpretedas a measureof the time to associateor substitute
letters and symbols. Other specific speedmeasureswere the
intercept and slope of the regressionequation relating reaction time to set size in a memory search task, the time to
reorganize information in memory, the time to name two
syllables,interferencein color naming, facilitation in color
naming, and the additional time required to count backwards
by small (3) or large (7) numbers.
One analytical procedurewith thesemeasuresconsistedof
comparingthe age-relatedvariancein the presumablyspecific
derived score before and after statistical control of the variancein the hypothesizedgeneralspeedfactor. Resultsofthis
procedure should indicate the extent to which age-related
influences on the measuresassumedto reflect specific processesare independentof more generalage-relatedeffects.
The second analytical procedure was designed to determine the degreeto which the age-relatedinfluenceson the
more complex measure,with the greaterprocessingrequirement, were independentof the age-relatedinfluenceson the
simpler measure.For example, the analyseswere used to
determine the degree to which the age-relatedinfluences in
the averageof the letter-symbol and symbol-letter reaction
time measureswere independentof those in the average of
the letter-letterand symbol-symbolreactiontime measures.
The rationalefor this procedurewas discussedby Salthouse
and Coon (1994), and it can be viewed as a means of
examining selectiveage-relatedeffects without the limitations associatedwith differencescoremeasures.
Finally, the relationsof the speedmeasuresto measuresof
memory were examinedin a structuralequationmodel and in
a series of hierarchical regression analyses. The primary
measuresof memory consisted of the number of items
recalledfrom free recall tasksinvolving l0 unrelatedwords.
In addition, accuracyvaluesin a letter memory task and in a
memory search task with interpolated activity between the
memory set and the probe also servedas secondarymeasures
of memory. It was anticipated that earlier results would be
replicated, such that there would be considerablereduction
of the age-related variance in the memory measuresafter
control of the measuresof speed.Of particular interest was
whether the reduction in age-related variance would be
greater with speedmeasuresassumedto representprocesses
relevant to memory functioning. Finally, in order to determine whether the relations to speedmight be greater when
the memory stimuli are presentedat a more rapid rate, three
different rates of stimulus presentationwere used in the free
recall memory task.
METHoD
Subjects. - Participants in this study consisted,of 172
adults between 18 and 93 years of age. (Data from 6
additional participants were omitted from the analysesbecauseof missingvalueson severalof the measures.)Demographic characteristicsofthe participants, divided into three
age groups for easeof description, are summarizedin Table
P3l
1. Notice that most participants rated their health betr.r'een
good (3) and excellent(l), and that the averageparricipanr
had attendedaboutone yearofcollege. Amount ofeducation
was negatively related to age, but at least one measureof
verbal ability, synonym vocabulary score, was positively
relatedto age.
Procedure.- All participantsperformedthe tasksin the
following order: number matching, pattern matching, pattern comparison, letter comparison, word copying. word
retrieval, vocabulary, Stroop color-word. naming speed.
counting backwards, sequential associates, free recall.
memorysearch,reactiontime, and serialmemory. Detailsof
the tasks are describedbelow.
The criterion task was afree recall memory task involving
two l0-word lists eachat stimuluspresentationratesof .5 . I .
or 2 secper word. After a practicetrial involving eight words
presentedat a 3-sec rate, the lists were presentedon a
computermonitor at ratesof 2 sec, I sec, .5 sec. .5 sec. I
sec, and 2 sec. At the end of the list the participanr*'as
allowed I min to write as many words as he or she could
rememberin any order. All lists consistedof tuo-srllable
words betweenfive and sevenlettersin lengthfrom Cluster8
(high in all dimensions)of the Toglia and Barrig ( 1978)
norrns.In addition, all words had a frequencl oi ar least20
per million according to the Thorndike tl9.l-lr r*ord frequency count. The primary measuresof pertbrmancewere
the numbersof items recalledat eachof the rhrec-presentation rates.
Several of the speed tasks were designedlo asressthe
speed of processespresumed to be directl\ rele\ant to
memoryfunctioning.One was a modifiedSrernbergmemory
search task involving letter stimuli in a ranc'd mapping
condition.Two blocksof 48 trials eachwere presenlc-d.
with
6 positive trials and 6 negative trials each lor ser-sizesI
through 4. The experimental trials were precc-dc'dby a
practiceblock of 8 trials, containing l trial ar eachcombination of set-sizeand decision type. During rhe Inrenal between the presentationof the memory set and the drsplayof
the probe stimulus a seriesof five randomll selecreddigits
(from 0 to 9) was presented.The researchpartlclpanrwas
instructedto pressthe "Z" key if the digit uas trtd and to
pressthe "1" key if the digit was even. and ro do rhis fbr
each of the interpolateddigits. The requiremenl1p ptrtorm
Table l. DemographicCharacteristics
ofResearchParticipants
(n : l12l
l8-39
Mean
n
Age
VoFemale
Education
Health
Vocabulary
Synonym
Antonym
*p < .ol.
69
28.r
63.8
13.9
1.9
4.0
3.'7
SD
40-59
Mean
60-93
SD
Mean
5.7
&
'to.4
8.2
'71.9
2.1
0.'7
39
48.0
64.1
13.6
2.2
2.O
0.9
2.7
2.'7
4.8
4.2
2.4
3.0
6.5
SD
13.0 2.7
Age r
- 10.
2.3
0.9
:.1.
6.0
.1 6
-1-1
-11
l{.
l0
P32
SALTHOUSE
five odd/even classification responsesduring the interval
betweenthe memory set items and the probe was designedto
maximize the involvement of secondarymemory in the task.
The reaction time and accuracyto the probe stimulus at each
memory set size were the primary measuresof performance,
although the slope and the intercept of the function relating
reaction time to the number of items in the memory set were
also computed.
A measureof rehearsalor articulation speedwas obtained
from the naming speedtask.ln this task two lists of l0 one"yllable or three-syllablewords were to be read as quickly as
possible.The lists were printed on a sheetof paper, and the
examinertimed the responseswith a stop watch. A1l of the
words were from Cluster 8 of the Toglia and Battig (1978)
norms and were arrangedin lists of l0 one-syllableand l0
three-syllable words. The measureof performance was the
average time required to name one-syllable and threesyllablewords (i.e., the total time divided by l0), which can
be hypothesized to serve as an index of articulation or
rehearsalrate.
The time required to reorganize information in memory
was determined from a serial recall task involving four
letters (Salthouse& Coon, 1993). Each trial in the task
consistedof the presentationof a list of four letters, and at
the end of the sequencean instructionindicatedwhetherthe
letters were to be recalled (by typing the letters) in the
original order or in alphabeticorder. Two blocks of l6 trials
eachwere presented,with eight trials in eachblock requiring
recall in the original order and eight trials requiringrecall in
alphabeticorder. The trials within a block were randomly
intermixed such that the researchparticipant did not know
about the order of recall until the presentationof the recall
instructionat the end of the list. A practiceblock of six trials
preceded the experimental trials. The primary measure of
performancein this task was the difference in the recall time
for correctlyrecalledsequences
in the two conditions,which
can be postulatedto reflect the time neededto carry out the
relevant (re-ordering or transformation) processing. The
number of trials with correct recall was also used as another
measureof memory elfectiveness.
Aword copying task required participants to copy a list of
words as rapidly as possible.Two lists of 24 words each
were presented,and 30 sec were allowed for the participant
to copy as many of the words as possible. The primary
measureof performance in this task was the mean time to
copy a singleword.
ln the word retrieval task the research participants read
lists of words and wrote simple associatesto those words as
quickly as possible. Each target word was preceded by a
common associateand had the first two letters followed by
blank lines indicating missing letters. The word pairs were
selectedfrom the middle-strengthassociatesprovided in the
appendixofNelson and Schreiber(1992). The target words
were the same as those used in the word copying task, and
thus at least some of the words may have been primed from
the previous task. The mean time per item in the task can be
hypothesized to reflect the time needed to gain access to
familiar material and write several letters of a word.
The sequential associates task involved the examiner
reading a target word and the researchparticipant producing
four successiveassociates.The examinerrecordedthe words
and also the total time required to generatethe four words. A
total of ten different target or stimulus words were used, all
of which were two-syllable words of high meaningfulness
(i.e., ratings5.27 to 5.98, mean : 5.49, from Clusters4
and 5 in Toglia & Battig, 1978).Five of the words were high
(5.14 to 6.10, mean : 5.76) and five were
in pleasantness
(2.38 to2.87, mean : 2.63) according
low in pleasantness
to the Toglia and Battig (1978) norns. (Associationtimes
were very similar, and highly correlated,r : .83, for the
high- and low-pleasantness
words, and thus the data were
collapsedacrossthis variable.) The measureof performance
was the time to generatefour successiveassociates,which
can be hypothesizedto reflectinternalassociationtime.
The principal criteria used to select other speed tasks
were: (a) minimal influenceof knowledge;(b) involvement
of a variety of responsemodes, stimuli, and processing
requirements;but (c) with little apparentdirect relevanceto
memory. Four of the nonmemory speed tasks were
computer-administeredreaction time tasks. In all casesthe
two-alternativechoice responsewas to a pair of stimuli
displayedin two vertically positionedboxesin the middle of
the screen. ln the letter-Ietter task and the symbol-symbol
taskthe stimuli were pairsof lettersor pairsof symbols,and
"/"
the decision was basedon physical identity, with the
"2"
pressed
key to be
if the stimuli were the same and the
key to be pressedif the stimuli were different. ln the lettersymbol and symbol-letter tasks the stimuli were lettersymbol or symbol-letterpairs and the decisionwas basedon
associationalequivalenceaccording to a code table presented at the top of the screen. lf the two stimuli matched
according to the code table then the " l" key was to be
"Z" key was to
pressed,and if they did not match then the
be pressed.Each task involved a practiceblock of l8 trials
and two experimentalblocks of 45 trials each. The tasks
were presentedin the order of letter-letter,symbol-symbol,
letter-symbol,symbol-letter,and then the sametasksagain
in the reverse order. The measureof performance was the
medianreactiontime in millisecondsfor the two blocks in a
given task. Averageaccuracywas greaterthan94Voin each
task and was not analyzed further.
Four tasks were based on paper-and-pencilprocedures.
The letter comparison and pattern comparison tasks were
identical to those used in earlier studies (e.g., Salthouse,
I 993b, 1994b),and involved the researchparticipant writing
"D"
"S"
between pairs of letters or
or the letter
the letter
pairs of line patterns according to whether they were the
same or different. Thirty seconds were allowed for each
page, and there were two pageswith each task.
The number matching and pattern matching tasks consistedof a page of rows with a target stimulus on the left and
four stimuli to the right, one of which was identical to the
target. The stimuli were two-digit numbers in the number
matching task, and simple patterns composed of geometric
figures in the pattern matching task. In both cases the
researchparticipant was to indicate stimuli that matched the
target by marking them with a circle or a slash.The measure
ofperformance in all four paper-and-pencilperceptualspeed
tasks was the number of items answeredcorrectly minus the
number of items answeredincorrectlv in 30 sec. In order to
AGE, SPEED,AND MEMORY
maintain comparability with the other measures,the scores
were divided into 30 to yield measuresin units of seconds
per item.
Two different vocal tasks were used to assessprocessing
speed. One was based on the Stroop color word task and
involved neutral (Xs), congruent(ink color matchesthe color
word), and incongruent (ink color is incompatible with the
color word) stimuli. Each condition consistedof four pages,
with 20 items on each page. The conditions were presented
in two counterbalanced sequencesat neutral, congruent,
incongruent,incongruent,congruent,and neutral.The measureof performancewas the mean time to name the colors in
the fourpagesofeach condition. Researchparticipantswere
instructed to correct errors when they occurred, but because
the correction time contributed to the total naming time,
errors were not analyzed separately.
The final speed task required the participants to count
backwards from a designatednumber by either I s, 3s, or 7s.
Four trials, each with a different starting number, were
presentedin each count back condition (i.e., by ls, 3s, or
7s), and the responseswere timed until five successive
numbers had been produced. The examiner monitored the
accuracy of the responses, correcting errors when they
occurred, and recorded the time to produce the five responses.The mean of thesetimes acrossthe four trials in
each condition served as the performancemeasure.
The vocabularytestconsistedof l0 five-alternativemultiple choicesynonymvocabularyitemsand l0 five-alternative
multiple choice antonym vocabularyitems. The items were
selected from published practice tests for the Scholastic
Aptitude Test, and were of moderate to high levels of
difficulty in samplesof collegestudents.Performanceon the
vocabularytestsconsistedof the numberof items answered
correctly in each section of the test.
RESULTS
Memory Measures
Descriptive statistics for the memory measuresiue presented in Table 2, and means in each of three age groups
expressedin standardizedunits are illustrated in the upper
left panel of Figure l. The recall scoreshad a maximum of
10, and the maximum value for the serial memory measure
was 8. Becausemany participants in the reordered (alphabetic) serial memory condition had scores of zero, this
measure was not used as an index of memory in later
analyses.
The estimated reliability was computed by determining
the partial correlation between the two scores for each
variable after controlling for age, and then boosting that
correlation by the Spearman-Brown formula. This yields
estimates of the reliability of the combined score that are
independentof the age-relatedeffects on the variable. The
reliability of a composite recall measureaggregatedacross
the three presentationrates was .68.
It is apparentin Table 2 that all memory measureshave
significant negative relations with age. Quadratic age trends
were also examined by entering the age-squaredterm after
the age term in a multiple regressionequation. Interactions
of age with gender, health, and education on the composite
P33
Table 2. Descriptive Characteristicsof Memory Measures
Variable
Recall - 0.5 sec
Recall - 1.0 sec
Recall- 2.0 sec
Memory search
Serial memory original order
Serial memory reordered
Mean
SD
Estimated
Reliability
Age r
.52
4.46
88.5
0.99
1.09
l .35
10.5
.51
.68
-.48*
-.42+
_.43*
-.26*
5.40
1.80
.74
-.25*
3.91
2.49
.89
-.25*
3.O2
3.s6
.) I
* p< . o l .
recall measureand the three other memory measureswere
also examinedin multiple regressionanalyses.None of the
quadraticage terms were significant, nor were any of the
interactions of age with gender, health, or education
significant.
A factor analysis(exploratorywith promax rotation) was
conductedon the threerecall measuresat different presentation times and the nonrecallmemory measures.Two factors
emergedin this analysis, with the first factor having high
loadingson the recall measuresand the secondfactor having
high loadings on the nonrecall (i.e., memory search and
serialmemory accuracy)measures.The correlationbetween
factors was .33, and the age correlationswere -.53 for the
first factor and -.29 for the second factor. The recall measureswith the three different presentationtimes loaded on
the samefactor and had similar correlationswith age, and
thus they appearto reflectsimilar processes
despitedifferent
absolutelevels of performance.Many of the later analyses
usedthe compositerecall measure(averageacrossall three
rates) to assessmemory becauseit is one of the simplest
measures,is most comparableto othermeasuresof memory,
and had the highestreliability of the recall measures.
SpeedMeasures
Descriptivestatisticsfor the speedmeasuresarecontained
in Table 3, and meanstandardizedscoresin eachofthree age
groupsareillustratedin the lower left, upperright, and lower
right panelsof Figure 1. (Becausemany participantswere
unable to perform the reordered serial memory task and the
counting backwards by 7 task correctly, the measuresfrom
these tasks were deleted from subsequentanalyses.)Estimated reliabilities, computed as describedabove, were generally high for the speedmeasures,with many of them above
.9. Also reported in Table 3 are correlations of the speed
measureswith age and with the composite recall memory
measure.All correlations were significantly different from 0
except for the correlation of age with the count back 3
measure.The quadratic age trends (age-squaredterm) were
significant with the pattern comparison (increment in R2 :
.05l), letter comparison(incrementin R2 : .063), number
matching(incrementin R' : .028), patternmatching(increment in R2 : .030), color neutral(incrementin R' : .056),
color congruent(incrementin R' : .069), and color incongruent (incrementin R' : .059) measures.In all cases,the
significant quadratic effect occurred becausethe age trend
P34
SALTHOUSE
Memory
Paper-and-Pencil
Speed
Pdem +Compadgqr
Rrcdl-.5
-
R€cCl-l
.a
N|'nb€rl/ffi*ng
R*dF2
""t""
E
o
o.2
8
o
t'it
o.2
7
Acc,
llemqv L FSercfi
-
*i;.\
t-cer_$4p*on
Pd6gM.dring
-..t
lUod-Cop)rirg
bte.MflDrYAcc.
Word
Rolrbvrl
-.ts.-
-t'2
!-'r
'N.
4.
VocalSpeed
IJ. RT
:
l{am. 1 s\dlaue
$S RT
tt rrr. O Sftt.U"
- - a - -
E
o
o.2
8
o
Fit
Ui RT
""o"..
Color Ndtrd
SL RT
Color ComDrliuc
....o.,..
/, . 2
libmow Seacfi RT
Odq lrrcornpcliH.
-..>..-
Se;rsrlid
L-etterMernry RT
Assoc.
Co|.|rnBad( 1
-
4.2
Oown Brd( 3
2 o g ) 4 0 5 0 q }
7 0 s )
Age
Ghronological
Figure l. Mean levels of performance, expressedin standarddeviation units, acrossthree age groups.
acceleratedin the oldestages.Age x healthinteractions,in
the direction of larger age relationsamong individuals with
low levels of self-reportedhealth, were significant for the
patterncomparison, letter comparison, memory searchreaction time, serialmemory original time, and the count backwards by ls and by 3s measures.No interactionsof age x
gender nor age X education were significant on any of the
speedmeasures.
Inspection of Table 3 reveals that the magnitude of the
relation to memory was similar for most of the speedmeasures. It is noteworthy that the relations to the composite
memory measurewere not appreciably higher for measures
hypothesizedto reflect processesthat might be particularly
relevant to memory (i.e., word retrieval, memory search,
serial memory, sequential associates)than for other speed
measures.
Multiple regressionequations were used to compute the
variance each variable shared with both age and recall
memory. First the R'for age alone was computed, and then
the increment in R2 for age after control of the speed
variable. Next, the latter value was subtracted from the
former, and divided by the former. When the product of
thesecomputationsis multiplied by 100 the result conespondsto the percentageof the age-relatedvariance(i.e., R' )
in the memory variable that is shared with the speed variable.As an example,the age-relatedvariancein the composite memory measurein the total sample was .289, and the
increment in R2 associated with age after control of the
pattem comparison measure was .084. Because (.289 y 00:7O.9Vo,it
. 0 8 4 ) 1 . 2 8:9 . 7 0 9 ,a n d . T 0 9 m u l t i p l i e d b 1
can be inferred that the age-relatedvariancein the composite
memory measurewould be reducedby 70.9Voif the variance
in the pattern comparison measurewere held constant.
Values in the last column of Table 3 indicate that most of
the speed variables shared at least moderate amounts of
variance with the age and memory variables. Once again,
however, it is important to note that the values are not
especially high for the measuresassumedto reflect processes
closely related to memory functioning. In fact, the measures
of word retrieval time and sequential associatestime had
relatively low estimates of shared variance with age and
memory.
AGE, SPEED,ANDMEMORY
P35
Table 3. Descriptive Characteristicsfor SpeedMeasures
.sD
Variable
Pattern Comparison
Letter Comparison
Number Matching
Pattem Matching
Word Copying
Word Retrieval
Letter-Letter RT
Symbol-Symbol RT
I€tter-Symbol RT
Symbol-lrtter RT
Memory Search RT
Serial Memory Orig. RT
Name - One Syllable
Name - Three Syllables
Color - Neutral
Color - Congruent
Color - Incongruent
Sequential Associates
Count Back - I
Count Back - 3
2.24
3.91
1.29
2.48
2.22
3.10
0.87
0.98
2.01
2.09
1.76
t . ' 7|
3 . 7|
5.20
ll.6'7
to.47
2t.16
17.68
5.59
16.47
Estimated
Reliability
0.95
1.95
0.39
1.33
0.58
I .39
0.43
0.57
o.g
0.70
0.65
0.93
o.94
1.25
3.99
4.09
7.5l
9.85
2.19
9.14
.7'7
.56
.89
.8t
.94
.81
.tJ
.82
.88
.94
.91
.93
.17
.87
.93
.91
.93
.96
.68
.85
Correlations
Age
.54*
.62+
.55*
.50*
.39*
.25*
.57*
.53+
.69*
.'72t
.62*
.56*
.53x
.46+
.51*
.65*
.21+
.l 8 *
.06
Memory
VoYanance
Shared With
Age and Memory
'70.9
_.598
-.48*
- . 5 1*
-.56*
-.51*
_.49+
- . 51 *
-.58*
-.54*
-.40*
-.394
_.40*
-.34*
-.30+
-.36+
-.31*
83.0
62.6
62.6
5 8 .I
38.4
65.7
65.l
87.5
88.2
61.2
72-3
49.5
42.9
47.4
38.8
75.8
18.3
19.7
6.9
*p < .ol
General - SpecificAnalyses
Structural equation analyses were the primary method
used to derive estimates of the general and specific agerelatedinfluenceson the observedspeedmeasures.An initial
exploratory factor analysison the 20 speedmeasureslisted in
Table 3 revealed four factors with eigen values greater than
one. However, the pattern of loadings of the variableson the
factors was not easily interpretable, and the correlations
among factors were moderatelyhigh (i.e., .37 to .58). A
confirmatory factor analysiswas next conductedwith factors
correspondingto paper-and-pencil(i.e., the first six variablesin Table 3), reactiontime (i.e., the secondsix variables
in Table 3), and vocal (i.e., the last eight variablesin Table
3) tasks. Although this model provided a moderatefit to the
; NFI = .90;
d a t a ( i . e . ,X ' [ N : l 7 2 , d f : l 4 l ) : 3 9 2 . 5 7 N
CFI : .92), the correlations between the factors were
extremely high (i.e., .77 between paper-and-penciland
reactiontime, .79 betweenreactiontime and vocal, and .97
between paper-and-pencil and vocal), suggesting that any
method-basedfactors that might exist were not very distinct.
On the basis of these preliminary analyses, the initial
model consisted of a single general speed factor and two
memory factors (corresponding to the recall and nonrecall
tasks). This measurementmodel provided a moderate fit to
the data after allowing correlated residuals between vari(i.e., X' lN : 172, df =
ablessharingthe sameassessment
:
:
.85; CFI : .87). The first
2641
723.07; NNFI
structural model addedage to the analysis, with a direct path
to the speedfactor and indirect paths, through speed, to the
memory factors.The fit of this model was adequate(i.e., X'
lN : r12, df : 2891: 800.46;NNFI : .85;CFI : .86).
Direct paths from age to all the observed speed variables
were then examined, and the pathswith coefficientsdifferent
from zero by two standarderrors retained. Six variableshad
significantcoefficients,and the fit of the resulting model,
portrayedin Figure 2, was somewhatbetter (i.e., X'[N :
1 7 2 , d f : 2 8 3 1 : 1 4 6 . 2 7 ;N N F I : . 8 6 ; C F I : . 8 8 ) .
Attempts to improve the fit of the model by adding direct
paths from age to the memory factors, and to the observed
memory variables, did not result in significant improvements in fit, or in any path coefficients that differed from
zero by more than two standarderrors. Becauseall relations
from age were examined,it can be concludedthat although
the fit of the model was not particularly impressive, it is
unlikely that age has more relations to the variables than
those already representedin the model.
The first point to note about Figure 2 is that most of the
age-relatedinfluenceson the individual speedvariablescan
be modeled as being mediatedthrough a generalor common
speedfactor. That is, the direct paths from age to the speed
variables are both few in number and small in magnitude.
Estimates of the general (mediated) age-related influences
on the variablescan be obtained from the product of the path
coefficientsfrom ageto the generalspeedfactor and from the
generalspeedfactor to the variable. Theseestimates,as well
as the estimates of the specific (direct) age-relatedeffects,
and the total age-related effects in the form of the age
correlation,are listed in Table 4.
According to path analysis logic, the correlation between
age and the variable can be partitioned into direct (specific)
and indirect (general) components. In other words, if these
are the only factors operating in this situation, then the sum
of the general and specific age-relatedeffects for a given
variable should equal the total age-related effects on that
variable. The correspondencesbetweenthe correlations and
the sum ofthe generaland specific coefficients in Table 4 are
P36
SALTHOUSE
.?
.74
Memory1
Fcoll-.5
Reall-l
tuall-2
Age ----z---->
Sneed (
-*\
.6 \
a07 \
-.12 \
110
\
patcrcm,,, .w
LcrCom
/,/
.U,
NumMlt
/'/
'n
PatMat
,/
,76
wRoPv ./'
.28)
'61
WF.t
..t
.51
LLrr
//
'8
./
.67
,11)
SSFT
LqRr
/"
'v
.8)
SLRT
//'
.Sg
MSEhRT
.rt
'71
SM-Or
a"
'73
,z
'm
Nlml
'35)
Namg
//
ct{ \
o]^...\
. t l
.m) I
d
-'23
\
/
'n
//'
'@
S.qANc
//
cnE"tl
,"
ut'nt"
3
.,,\ su-occ
'6
///
//
M"toryz
.83
's
'g
-rs \. fA"is ,./ .u
Figure 2. Resultsof structuralequation model analysiswith speedand
memory measures.The memory variables are listed in Table 2 and the speed
variables in Table 3. Lines to the sp€edvariables representpaths from Age
or from the General Speed factor, and lines to the memory variables
representpaths from the Memory factors. All coefficients are standardized.
not exact, but they are close enough to suggest that the
general and specific estimates from this analysis may be
reasonableapproximationsto the true values.
A similar analysis was carried out when the first (unrotated) principal component from a principal components
analysis was used as the estimateof the general speedfactor
[seeJensen(1980)and Ree & Earles(1991)for the rationale
underlyingthis method of estimatinga generalfactorl. The
standardized regression coefficients in a simultaneous regressionequation with the first principal componentand age
as predictors of the speedvariable were then used to derive
estimatesof the general and specific age-relatedinfluences.
That is, the specific effect was the regression coefficient
from age (when it was significantly, at p I .05, different
from zero), and the general effect was the product of the
regressioncoefficient from the first principal componentand
the correlation coefficient (i.e., .69) between age and the
first principal component. These estimatesderived from the
principal componentsanalysisare listed in the fifth and sixth
columnsof Table 4.
Although the values of the estimatesdiffer according to
how the general speed factor is derived, it is clear that for
most of the variablesa very large proportion of the total agerelated variance was mediated through the general speed
factor. The generalspeedeffect was larger than the observed
age effect on severalmeasures(e.g., word retrieval, counting backwards by ls, and counting backwards by 3s),
suggesting that other factors may have been operating to
counteract the general influence. Factors operating to increasethe size of the age effects may have been operating
when the estimatesof the specific effects are the samesign as
those for the general effects (e.g., the symbol-letter RT
measurein the principal componentsanalysis).
The inference that a considerable amount of the agerelated variance on the speed measureswas shared is also
confirmed by analyses of quasi-partial correlation coefficients (Salthouse, 1994a). Unlike traditional correlations,
which reflect the proportion of the total variance that is
shared between two variables, quasi-partial correlations
reflectthe proportionofage-relatedvariancein two variables
that is shared.The median quasi-partialcorrelationamong
the 19 speed variables (excluding the count back by 3s
measurewhich had very little age-relatedvariancethat could
be partitioned)was .79, with a rangefrom .45 to .99. It can
therefore be inferred that, on the average, approximately
(i.e., .79' ) of the age-relatedvariancein pairs of speed
62%;o
variableswas shared.
Derived Measures
In addition to the directly observedspeed measures,a
numberof derived measurescan be createdto representthe
duration of potentially more specific processes.These derived measuresaredescribedin Table 5 along with summary
statisticsfor the measures.Each measureis obtainedeither
from subtraction or from a linear regressionequation, and
they can be interpreted as representingthe time needed to:
(a) searchthe code table and substituteitems;(b) searchand
retrievean item from memory; (c) reorderitems in memory;
(d) articulate(and possibly rehearse)verbal material; (e) suppress incongruent or incompatible information; (f) benefit
from congruent or compatible information; or (g) carry out
easy and difficult mental manipulations.
Many of the derived measureshave moderate reliability,
and somealso have moderatecorrelationswith the composite
memory measure.However, the correlations are not higher
for the measurespresumedto be most relevant for memory
functioning than for the other measures.To illustrate, the two
derivedmeasuresfor which the correlationswith the composite memory variable are highest are substitutionand interference, which are not obviously related to memory. Furthermore, the interceptof the memory searchfunction has higher
relations to age and to the composite memory variable than
does the slope parameter,despite a general assumptionthat
the intercept reflects primarily perceptual and motor processeswhereas the slope reflects processesassociatedwith
searchthroughinformationin memory.
The independence of the age relations in the derived
measureswas examined with two analytical procedures. In
AGE. SPEED.AND MEMORY
P37
Table 4. Estimatesof General and Specific Age-Related Influences on SpeedMeasures
Variable
.54
.62
.55
.50
.39
.25
.57
.53
.69
.72
-62
.56
.53
.46
.51
.54
.65
.21
.18
.06
Pattern Comparison
Letter Comparison
Number Matching
Pattern Matching
Word Copying
Word Retrieval
Letter-Letter RT
Symbol-Symbol RT
Letter-Symbol RT
Symbol-Letter RT
Memory Search RT
Serial Memory Orig.
Name - One Syllable
Name - Three Syllables
Color - Neutral
Color - Congruent
Color - lncongruent
SequentialAssociates
Count Back - I
Count Back - 3
PCA Estimates
SEM Estimates
Total Age Effect
Correlation
General
Specific
.62
.63
.61
.59
-.06
0
0
-.07
.4t
_ 1 )
.39
.52
.52
.63
.&
.57
.56
.54
.51
-.10
0
0
0
0
0
0
0
0
0
0
0
0
-.23
-.28
.)f
.52
.64
.35
.42
.34
General
Specific
.5'7
.51
.59
.55
.49
.50
.40
.45
.46
.42
.40
.46
.52
.53
0
0
0
0
0
1A
.tI
0
-zJ
.29
.22
0
0
0
0
0
0
-.26
.)l{
.50
.55
.48
.60
^ 1
--)lt
Note: Specific effects identified when p < .05
Table 5. Descriptive Characteristicsof Derived SpeedMeasures
Variable
Substitution Time
(I-etter-Symbol + Symbol-Letter RT)
- (Letter-ktter RT + Symbol-Symbol RT)
Memory Search Slope
(Slope of regressionof memory
search RT on number of memory set items)
Memory Search lntercept
(lntercept of regressionof memory
search RT on number of memory set items)
Memory Reorganization Time (n : 148)
(Time to recall letters in alphabetic
order minus time to recall letters in original order)
Articulation Time
(Time to read 3-syllable words minus
time to read l-syllable words)
Stroop Interference Time
(Time to name incongruent colors minus
time to name neutral colors)
Stroop Facilitation Time
(Time to name neutralcolors minus
time to name congruent colors)
Easy Count Back
(Time to count backwards by 3 minus
time to count backwards by l)
Difficult Count Back (n : 161)
(Time to count backwards by 7 minus
time to count backwards by l)
*p < .01
Mean
SD
Estimated
Reliability
Conelations
Age
Memory
VoYaiance
Shared With
Age and Memory
-.50*
.90
3.1'7
1.04
0.06
0.12
02
.o2
1.60
o.'72
.524
-.39*
47.'7
1.36
1.08
.23*
-.34*
r8 . 0
1.49
o.'73
.10
-.t4
4.5
9.50
4 . 7|
.60*
-.49+
1.20
1.89
r0.88
8.29
23.53
t7.63
.39
-.09
.79
.03
-.01
-.10
1a*
-.17
0
6't.8
0
2.8
0
I
P38
SALTHOUSE
the first method, the variance in the estimate of the general
speedfactor was controlled before examining the age-related
variance in the derived measure. The estimates of agerelated variance before and after control of the variance in
the general speedfactor (basedon the first principal component) are summarized in the second and third columns of
Table 6. It can be seen that for most variables there was a
considerablereduction of the age-relatedeffects in the presumably specific measuresafter control of the general speed
factor. The derived measuresfrom the counting backwards
tasksare notableexceptionsbecausewith thesemeasuresthe
age-relatedeffects actually increased when an estimate of
the general factor was controlled.
The secondanalyticalmethodconsistedof controllingthe
variancein the simpler measure(e.g., choice reactiontime
with the same stimuli, intercept of the memory search
function) when analyzing the age-related variance in the
complex measures(e.g., choice reactiontime with stimuli
associatedaccording to a code table, slope of the memory
searchfunction). The estimatesofthe age-relatedvariancein
the complex measurebefore and after control of the simple
measure are presented in the fourth and fifth columns of
Table 6. Notice that there was a decreasein the age-related
variancefor most measures,althoughsignificantuniqueagerelated variance remained for the substitution, interference,
and facilitation measures.It can also be seenthat statistical
control of the intercept variable resultedin an increaseof the
age relations for the slope measure.A similar outcome was
alsofound by Salthouseand Earles(1995), and it is probably
a consequenceof a suppressorrelation in which age is
positively relatedto the intercept(r : .54), but the intercept
is negativelyrelatedto the slope(r -- -.45).
The major implication of the analysessummarizedin
Table 6 is that many of the age-relatedinfluences on measurespresumedto reflectspecifictypesofprocessingare not
independentof the age-relatedinfluenceson other measures
of speededperformance. Some unique age-related effects
are evident in the substitution and interferencemeasures,but
even thoseeffects are only a fraction the size of the total agerelated effects on the original variables.
SpeedMediation of Age Dffirences in Memory
The structural model portrayed in Figure 2 does not
contain any direct paths from age to either the memory
factors or the memory variables. The implication is that all
of the age-relatedvariance in thesemeasuresof memory was
mediated through age-related effects in the general speed
factor.
Another set of analyses was conducted to explore this
implication further. For this purposethe first principal component from the principal componentsanalysis was used as
the estimate of the general speedfactor, and the composite
recall memory measure, expressedin standard score units,
was used as the primary memory measure.Relations among
thesemeasuresand age are illustrated in Figure 3. Note that
there were moderate relations between age and speed, between age and memory, and betweenspeedand memory, but
a very weak relation between age and memory after control
of speed. An additional analysis revealed that the age x
speedinteraction was not significant in the prediction of the
Table 6. Age-related Variance in Derived SpeedMeasures
Derived Measure
Construct
Substitution Time
Memory Search Slope
Memory Search Intercept
Memory Reorganization
Time (n : 148)
Articulation Time
Stroop Interference Time
Stroop Facilitation Time
Easy Count Back
Difficult Count Back
(n = 16l)
Comolex Measure
R2
Age
R2 for Age
After General
.421*
.000
.291*
.M2*
.005
.012
.504*
.000
.291*
. l4l *
.097*
.054*
.01I
.361*
.008
.001
.n2
.010
.o24*
-026
.128*
.150+
.2to*
.419*
.2904
.004
.01I
.001
.066*
.010*
.005
.000
.120*
.m0
.005
R2
Age
R2 for Age
After Simple
* p< . o l .
composite memory variable, and thus there is no evidence
that the speed-memoryrelation varied as a function of age.
Statistical control results with other measuresof memory
and speedare summarizedin Table 7. It is apparentthat the
pattern was very similar for the composite recall measure,
and a composite of accuracy scores in the memory search
and serial letter (original order) memory tasks. In all cases
the largest reduction of age-relatedvariance occurred with
control of the general speed factor. The reduction of agerelated variance was nearly as large with the perceptual
speed composites, but there was much less reduction after
control of the speedmeasuresthat might be hypothesizedto
reflect memory-relevant processes.
DIScUSSIoN
The methods used in this study to estimate the general
speed factor are based on somewhat different assumptions
and on different types of variance. For example, in the
structural equation method and in the principal components
analysis the total variance in the variables was analyzed,
whereas in the quasi-partial analyses only the age-related
variance was examined. However, all three methods are
basedon the lack of independence
amongvariables.That is,
each method is basedon the premise that if the measuresare
independentthen they should be uncorrelated, but if they
have moderateto large positive intercorrelationsthen they are
consistentwith the existenceof a common, or general,factor.
The resultsof the three analytical methodsare consistentin
indicating that many speed variables have a considerable
amount of common or sharedage-relatedvariance. It therefore seems meaningful to refer to a general speed factor
becausethe variables are not independent,and the existence
of a general factor is a parsimoniousway to account for this
lack of independence.It should be emphasized,however,
that the inference of a general factor does not imply that the
agetrendson the measuresneedbe identical. That is, the age
relations on different measuresare not necessarilyof equivalent magnitude becausethe variables can differ with respect
to their loadings on the general factor, and because some
variables could also have direct (or specific) relations from
age that serve to moderate the age-relatedeffects mediated
AGE, SPEED,ANDMEMORY
A
P39
B
3
2.5
2 R2= ,48O
1.5
1
ta.t
a
t
a
. .
!.
l
J
.
-
a
a ,
a a
t a
a a
a
.
a
20 30 .O 50 60 70 80
ChronologicalAge
t
c
\ a
a a
- a
a
a
a
\
a a a a
-
.
a
a
-
0
Speed
D
t
.
a
a
a
a
a
-
..
^2
R-= .289
a
a
a
a
a a
9
a
a
a
a
a
a a
a
a
o
b
E
o
=
a
o
a
a
a a
aaaa
aaa
a
a
aa
a
a
a
3
2.5
2 a'a
1.5
1 !.^.
0.5
R2=.014
a
t
'
tat
r
.
(
a
.
r
o
'l o a oo '
.1.
:
....1
i-.' 1:
-.' . .
€
b 4.5
o -1
E
o -1.5
.
.l'i, i!...:"i;t1T
- 3
r - l t .
.
a
"
t
t
a ; t
rooo
o
a
t
.
a
-2
-2.5
a
3
ChronologicalAge
20 30 ,+0 50 60 70 80 90
ChronologicalAge
Figure 3. Relations between age, the principal components analysis estimateof the general speedfactor, and the composite recall memory measure. Both
speedand memory are scaled in standardizedunits wilh a mean of0 and a standarddeviation of l.
Table 7. Age-relatedVariance in Memory
Before and After Control of Speed
Measures of
Memory
Other
Memory'
Age Alone
Age After Control of:
General SpeedFactor
PerceptualSpeed I
(Letter Comparison and Pattem Comparison)
PerceptualSpeed2
(Number Matching and Pattern Matching)
Word Retrieval
Sequential Associates
Memory Search Slope
Memory Reorganization Timeb
Articulation Time
.289*
.016
.035*
.005
.084+
.178*
.236*
.290*
.192*
.276*
.o2l
.067*
.059x
.091*
.035
.087*
"Based on accuracy in Letter Memory (original order) and Memory
Searchtasks.
bn : 148.
*p <'ol.
throughthe generalfactor. To illustrate,relativelysmall age
effects were apparenton the word retrieval measure,perhaps
becausethat measurehad a positive relation with the measures of word knowledge which tended to be greater with
increasedage (also see Salthouse, 1993b, for a similar
finding).
Becausethe inference that for many speedvariables most
of the age-relatedvariance can be assumedto be mediated
through a generalspeedfactor is rather surprising, data from
three earlier projects in my laboratory were also analyzedby
these methods. One data set (described in Earles &
Salthouse, 1995) is based on three separatestudies and
consists of data on two reaction time and four paper-andpencil speedmeasuresfrom each of744 adults. The second
data set consistsof data on I I paper-and-pencilspeedmeasuresfrom eachof 305 adults(Salthouse,I 993a).Data in the
third data set consist of two reaction time, four paper-andpencil, and l0 vocal speedmeasuresfrom eachof242 adults
(Salthouse& Meinz, 1995).
Estimates of the general and specific age-related influences derived from the structural equation and principal
componentsanalysismethodsare listed in Table 8. Note that
P40
SALTHOUSE
Table 8. Analyses of Data from Other Studies
Variable
Total Age Effect
Correlation
SEM Estimates
General
PCA Estimates
Specific
General
Specific
0
.10
0
0
0
0
_.49
-.51
_.49
-.51
_.47
-.48
-.08
.10
0
0
0
-. l0
-.il
-. l0
0
0
0
0
0
0
0
.13
.20
_.40
-.40
-.45
-.43
_.44
-.49
_.49
-.43
-.44
-.43
-.50
-.19
-.10
0
0
0
0
.08
-.10
-.1I
.19
.28
.12
.13
.18
.t4
0
0
-.09
0
0
0
0
0
0
0
-.05
-.06
.38
.26
.22
.26
Earles & Salthouse(1995J. N : 744
Pattern Comparison
lrtter Comparison
Boxes
Digit Copy
Digit Digit RT
Digit Symbol RT
-.56
-.41
Pattem Comparison
Letter Comparison
Digit Symbol
Horizontal Marking
Vertical Marking
Letter Copy
Number Copy
ktter Completion
Number Completion
l,etter Transformation
Number Transformation
-.59
-.51
-.51
-.38
-.40
-.49
-.41
-.53
-.55
-.58
-.51
^a
/a
-.46
-.50
-.58
-.48
-.47
-.55
Salthouse(1993a),N : 305
-.48
_.47
-.51
_.40
-.43
-.51
-.53
-.53
-.51
-.38
-.44
-.23
Salthouse& Meinz (19951,N : 242
Pattem Comparison
Letter Comparison
Digit Digit RT
Digit Symbol RT
Symbol Copy
Digit Copy
Reading Speed
Color - Compatible
Color * Neutral
Color - Incompatible
Position - Compatible
Position - Neutral
Position - Incompatible
Number - Compatible
Number - Neutral
Number - Incompatible
.63
.56
.59
.63
.48
.45
.2'l
.48
.50
.57
.49
.45
.44
.55
.48
.48
.45
.31
.40
.45
.34
.39
.44
.51
.50
.51
.54
.53
.51
. )/
.56
.56
.J+
.JJ
.38
.42
.41
.55
.54
.53
.50
.58
.60
.58
.57
.61
.60
1A
0
0
_)1
0
0
0
-.09
-.15
-.t4
0
-.14
t a
Nores.Fitsof thesinglespeedfactorstructuralequationmodelwere:Earles& Salthouse
( 1995):y, IN : 744,df : I I ] : 30.61;NNFI : .9g;CFI : .99;
: 4, d
5 f1 : 2 3 3 . 9 0 . N N F I : . 9 2 ; C F I : . 9 5 ; S a t t h o u s e & M e i n z ( t 9
Salthouse(1993a):1r[N:305
9 [5N) :: 2 4 2 , d f : 1 0 4 ]= 4 6 4 . 0 5 ; N N F I :.89;
y2
CFI= .91.
for every variable a large proportion of the age-related
effects can be inferred to be general, rather than specific.
Results of the quasi-partial correlations are also consistent
with this inferenceas the medians(andranges)were .83 (.70
to .94) in the Earlesand Salthouse(1995) data, .86 (.69 to
.99) in the Salthouse( 1993a)data,and .36 (.66 to .99) in the
Salthouseand Meinz (1995) data.
There is also little evidencein the current analysesthat
the derived measuresare distinct from the general factor.
Measures based on difference scores or linear regression
slopes are often postulated to reflect specific cognitive
processesbecauseother processesare presumedto be eliminatedor subtractedout when the derived score is computed.
However, the results in Table 7 indicate that a large proportion of the age-related variance in the difference score
measuresis shared with the inferred general speed factor.
Furthermore, much, but not all, of the age-relatedvariance
in the more complex measures is shared with that in the
simpler measures.
These results have important implications for research
focusing on measureshypothesizedto reflect the duration of
specific cognitive processessuch as rehearsal, access of
information in memory, and inhibition of irrelevant information. That is, while measuressuchas thesemay representthe
hypothesized processesin a sample of age-homogeneous
adults (althoughconverging evidencefrom other experimental procedureswould clearly be desirable), the results ofthe
current analysessuggestthat large proportions of the agerelated variance in these types of measuresare shared with
other measuresof speededperformance. Efforts to obtain
precise measures of specific cognitive processes should
therefore proceed with the recognition that those measures
AGE. SPEED.ANDMEMORY
may have little or no unique age-related variance. At the
very least, researchersinterestedin investigating age-related
influences on measurespostulated to reflect the duration of
specific cognitive processeswould be well advised to use
analytical methodsto establishthat the age-relatedeffects on
those measuresare independentof the age-relatedeffects on
other measuresof processingspeed.Without such evidence,
a very plausible interpretation of the age-relatedeffects on
the purportedly specific measures is that they are merely
another manifestation of the negative relations between age
and a general speedfactor.
The correlation-basedmethods describedabove appearto
have several advantagesover the method of examining systematic relations (i.e., Brinley plots) for investigatingthe
existenceof a general speed factor (e.g., Cerella, 1994;
Myerson, Wagstaff, & Hale, 1994;Salthouse,1992). First,
the variablescan be in different units (e.g., reactiontime,
number correct in a timed test) and need not be expressedin
the same measurementscale, as is required when performanceof older adults is plotted as a function of performance
of young adults. Second,these methodsyield quantitative
estimatesof the relative contribution of general and specific
influences,rather than all-or-none claims regardingthe existenceof a generalspeedfactor. Unproductiveargumentsabout
whether all of the age differences can be attributed to a
generalspeedfactor can thereforebe replacedwith objective
indicatorsof the relative influenceof different types of determinants, in the form of estimatesof the proportions of agerelated variance. And third, the relation of the inferred general factor to other aspectsof cognition can be examinedwith
correlation-basedprocedures.This last property is important
becausespeedper se, is often of limited interest, and instead
is primarily interesting with respectto its causesand consequences,and particularly its relations to variablesreflecting
memory and other aspectsof cognitive functioning. In this
regard, the report by Salthouse(1993a) that the slope of the
systematic relation function for individual researchparticipants was not very highly related to measuresof cognitive
functioning raises serious questions about the meaning of
measuresderived from systematicrelation functions.
The results of this study replicate findings of several
earlier studiesindicating that speedmeasuressharedconsiderablevariancewith ageand with measuresof memory. One
new finding in the current study is that the greatestreduction
in age-relatedvariance occurred with control of the general
speed factor. There was also a strong reduction of agerelated variance with the perceptualspeedmeasures,which
had high loadings on the generalfactor. However, there was
only a weak reduction of the age-related variance with
measuresthat could be hypothesizedto reflect the duration of
processes specific to memory. The results of this study
thereforehelp in clarifying the nature of the speedmediation
of adult age differences in memory. That is, the mediation
doesnot appearto occur simply becauseof a reduction in the
speedof memory-specificprocesses,but insteadseemsto be
a reflection of age-relateddifferences in a very broad speed
factor that is most clearly indicated by measuresof how
quickly many elementary cognitive operations can be executed. Future researchshould be directed at investigating the
causesof the age-related differences in this general speed
P41
factor and at explicating the mechanismsby which the rate at
which cognitive operations are executed affects the quality
of the products of cognitive processing.
AcKNowLEDGMENTS
This researchwas supportedby NIA Grant AGR37 06826. I would like to
thank Stephen Camp, Nate Fristoe, Brian Judd, George Manning, Betsy
Meinz, Soo Rhee, Dwana Starks, and Jocelyn Thomas for assistancein
recruiting participants and collection of data, and the valuable suggestions
of two anonymous reviewers.
Address correspondenceto Dr. Timothy A. Salthouse, School of Psychology, Georgia Instituteof Technology,Atlanta GA 30332-107O.
REFERENcES
Bors, D.A., & Fonin, B. (ln press).Age, speedof informationprocessing'
recall, and fluid intelligence.Intelligence.
rate. [n J.E. Birren &
Cerella,J. ( I 990). Aging and information-processing
K.W. Schaie (Eds.), Handbook of the psychoktgt of aging (3rd ed.)
(pp.2Ol-221). San Diego: Academic Press.
Cerella, J. (1994). Generalizedslowing in Brinley plots Journal ofGerontology: Psychological Sciences,49, P65-P'7l.
Craik, F.l.M.. & Jennings,J.M . (1992).Humanmemory' In F.LM. Craik &
T.A. Salthouse(Eds.), H andbook of aging and cognit ion (pp. 5 I - I I 0).
Hillsdale,NJ: LawrenceErlbaumAssociates.
Earles,J.L., & Salthouse,T.A. (1995). Interrelationsof age, health, and
speed. Journal of Gerontoktgy: Psychological Sciences, 508,
P33-P41.
Hultsch, D.F., & Dixon, R.A. (1990). Leaming and memory in aging. In
J.E. Binen & K.W. Schaie (Eds.), Handbook of the psychology of
aging (3rd ed.) (pp. 259-2'14). San Diego: Academic Press.
Hultsch, D.F., Hertzog, C.K, & Dixon, R.A. (1990). Ability correlatesof
memory performance in adulthood and aging. Psychology and Aging, 5,
356-368.
Jensen,A.R. (1980). Eias in mental resling.New York: The Free Press.
Kausler, D.H. (1994). Learning and memoryin normal aging. San Diego:
Academic Press.
"ShiftKliegl, R., & Mayr, U. (1992). Commentaryon Salthouse(1992)
ing Levels of Analysis in the Investigation of Cognitive Aging."
Human Development, 35, 343-349.
McArdle, J.J., & Prescott,C.A. (1992). Age-basedconstructvalidation
using structural equation modeling. Experimental Aging Research, 18,
87-l 15.
Myerson, J., Wagstaff, D., & Hale, S. (1994). Brinley plots, explained
variance, and the analysis of age differences in reaction latencies.
Journal of Gerontology : P sychological Sciences,49, P72-P80.
Nelson, D.L., & Schreiber,T.A. (1992). Word concretenessand word
structure as independentdeterminantsof recall. Journal of Memory and
Language,3 1, 23'7-260.
Park, D.C., Smith, A.D., Lautenschlager,G., Earles, J., Frieske, D.'
Zwahr, M., & Gaines, C. (1994). Mediators of long,-termmemory
performance across the Iift span. Manuscript submitted for publication.
Ree, M.J., & Earles,J.A. (1991). The stability of convergentestimatesof
g. I ntelligence,I 5, 2'7| -2'18.
Salthouse,T.A. (1985). Speedof behavior and its implicationsfor cognition. In J. E. Birren & K.W. Schaie(Eds.), H andbook of thepsychology
of aging (2nd ed.). New York: Van NostrandReinhold.
Salthouse, T.A. (1991). Theoretical perspectiveson cognitive aging.
Hillsdale, NJ: Lawrence Erlbaum Associates.
Salthouse.T.A. (1992). Shifting levels of analysisin the investigationof
cognitive aging. H uman D evelopment, 35, 321-3 42.
Salthouse,T.A. (1993a). Speed mediation of adult age differences in
cognition. DevelopmentalP sychology,29,'122-'738.
Salthouse,T.A. (1993b). Speedand knowledge as determinantsof adult
age differences in verbal tasks. Journal ofGerontology: Psychological
Sciences.48.P29-P36.
Salthouse, T.A. (1994a). How many causes are there of aging-related
decrements in cognitive functioning? Developmental Review, 14,
413-43'7.
Salthouse,T.A. ( I 994b). The nature of the influence of speedon adult age
differences in cognition. Developmental Psychology, 30' 240-259.
Salthouse,T.A. (1994c). Aging associations:influenceof speedon adult
P42
SALTHOUSE
age differences in associative learning. Journal of Experimental psychology: Learning, Memory, and Cognition,20, 1486-1503.
Salthouse,T.A. (1995). Differential age-relatedinfluenceson memory for
verbal-symbolic information and visual-spatial information. Journal oJ
Gerontology: Psychological Sciences,508, Pl93-P201
Salthouse,T.A., & Babcock, R.L. (1991). Decomposingadult age differencesin working memory. DevelopmentalP sychology,27, 76j-77 6.
Salthouse,T.A., & Coon, V.E. (1993). Influenceoftask-specificprocessing speed on age differences in memory. Journal of Gerontology:
Psychological Sciences,48, P245-P255.
Salthouse, T.A., & Coon, V.E. (1994). Interpretation of differential
deficits: The caseof aging and mental arithmetic. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, ll'1.2-1182.
Salthouse,T.A., & Earles, J.L. (1995). Age, health, and specific and
nonspecific measures of processing speed. In A.M. Ferrandez & N.
Teasdale(Eds.), Changes in sensory-motor behavior in aging. Amsterdam: Elsevier.
Salthouse,T.A. , Fristoe,N., & Rhee, S. (In press).How localizedareagerelated effects on neuropsychological measures?Neuropsychology.
Salthouse,T.A., & Meinz, E.J. (1995). Aging, inhibition, working memory, and speed,.Journal ofGerontology: Psychological Sciences, 508,
P29'7-P306.
Thorndike, E.L. (1944). The teacher's word book of 30.0A0 words. New
York: Columbia University Press.
Toglia, M.P., & Battig, W.F. (1978). Handbookof semanticwoid norms.
Hillsdale, NJ: LawrenceErlbaum Associates.
ReceivedFebruary 20, 1995
AcceptedJune 9. 1995