J ournal of Geronrolog : PSYCH OLOCIC AL SCI ENC ES
1995. Vol. 508. No. 6, P297-P306
Copyright 1995 br The Gerontological Stiety ofAmerica
Aging,Inhibition,WorkingMemory,andSpeed
Timothy A. Salthouseand ElizabethJ. Meinz
School of Psychology, Georgia Institute of Technology.
An implication of the hypothesis that failures of inhibition contrtbute to adult age dffirences in working memory
(Hasher & Zacks, 1988) is that statistical control of measures of inhibition should reduce the oge-rclated effects on
working memory. This implication was tested in a study in which interference rneasuresfromthree variants of a Stroop
task semed as the measures of inhibition. Although the age-related variance in measures of working mimory was
substantially reduced after control of the interference ,neasures, the degree of attenuation was at leasl as large when
speed measures from other tasks were controlled. Furthermore, additional analyses revealed that speed measures
from tasks requiring oral, written, and keypress responses shared large proportions of their age-relnted variance, It
wcs sugges(cd lhai sge-relobd inSuentes on sper!flr prDDessEr, smh as inhibilion, cannotbe accurately cssesse,l
unless the contrtbufions of more general age-related influences are taken into consideralion
THERE has recentlybeenconsiderableinterestin inhibir tion as a factor that might contribute to adult agerelated effects on cognition. For example, Hasher and
Zacks (1988) suggestedthat age differences in working
memory functioning originate becauseof reduced inhibition abilities. According to theseauthors, either irrelevant
material gets into working memory and lowers its functional capacity, or off-track material within working memory is not inhibited and distraction occurs during processing. Because working memory is postulated to play a
critical role in many cognitive tasks, a mechanismsuch as
this has the potential to explain the age-relatedeffects in a
wide varietyof cognitivemeasures.
An implicationof the Hasherand Zacksview is that if one
could control the individual differencevariationin a measure
of inhibition, then the age-relatedeffectson working memory, and on othermeasuresofcognition, shouldbe considerably reduced.(SeeSalthouse,1992a,for a discussionofthe
rationale underlying the application of statistical control
methodsin researchon aging and cognition.) Although the
statistical proceduresfor controlling the variance in one
variable when examining the relation between two other
variables are fairly straightforward, the challenge in the
applicationof theseproceduresin the current context is in
identifying adequatemeasuresof the inhibition construct.
Severalrequirementscan be identifiedfor suchmeasures.
One is that the measureshould be reliable, becauseif a
measurehas little systematicvariancethen it is unrealisticto
expect much of the total variance in the measureto be shared
with other measures.The measureshould also be valid in
order to have confidence that it actually reflects the target
construct- in this caseinhibition, rather than some other
construct. This often must be determinedfrom the pattern of
correlations with other measureshypothesizedto reflect the
same construct becausethere is currently no external criterion againstwhich one can validatemeasuresof the inhibition construct. Both reliability and validity benefit if the
constructis assessed
with multiple measures.Not only is the
reliability greaterdue to aggregation(Rushton, Brainerd, &
Pressley, 1983), but the validity is often higher because
constructvariancecan be emphasizedby cancellingout the
specific variance associatedwith the particular methods,
materials,and measures.
Severalalternativemethods have been used to measure
inhibition, but nearly all are indirect, and the reliability of
someof the measurescan be questionedbecausethe effects
are often quite small (e.g., measuresof negative priming
sometimesusedas an index of inhibition often averageonly
10-15 msec). The potentially low reliability may have
contributedto the lack of strong correlationsamong measurespresumedto assessinhibition in earlier studies(e.g.,
Hartman& Hasher, I 99 I ; Stolzfus,Hasher,Zacks,Ulivi, &
Goldstein,1993).
Inhibition is assessedin this study with three variantsof
the Stroop interferencetask. The dominant interpretationof
interferencein this task is that the slower responseto name
the color of a targetword when the identity of the word is in
conflict with the color occursbecausethe word information
is activatedautomaticallyand the researchparticipantneeds
to inhibit the irrelevant(word) informationwhen attempting
to namethe color (e.g., Cohn, Dustman,& Bradford, 1984;
Dulaney& Rogers, 1994;Macleod, l99l). Ir may therefore
be possibleto use the amount of interferencein the task as an
index of the effectivenessof inhibition, with larger interferenceinterpretedas a reflectionof decreasedinhibition.
If this interpretationof the interferencemeasureis correct,
and if older adultshave reducedinhibitory abilities relative
to younger adults, then the amount of interference in the
Stroop task would be expected to increase with increased
age. In fact, there are numerous reports of greater interferencewith increasedage in Stroop tasks(Cohn et al., 1984;
Comalli, Wapner, & Werner, 1962;Daigneault,Braun, &
Whitaker, 1992; Dulaney & Rogers, 1994;Hartley, 1993;
Hartman& Hasher,l99l;Houx, Jolles,& Vreeling, 1993;
Panek,Rush,& Slade,1984).
However, several issues need to be considered before
these results can be accepted as indicating an age-related
decreasein inhibitory ability. First, greaterinterferencewith
increased age could originate either because of reduced
inhibition and equal strength of automatic activation or
P297
P298
SALTHOUSEANDMEINZ
becauseof equal inhibition and increased strength of automatic activation.That is, insteadof older adultshaving less
effective inhibition processes,their greaterinterferencemay
occur becauseof more potent or effective automatic activation such that the irrelevant word information is activated
more strongly in older adults than in young adults. One way
in which thesepossibilitiesmight be distinguishedis by also
measuringfacilitation, or the benefit of compatible word and
color information relative to a neutral condition. Facilitation
should reveal any benefits of automatic activation, and thus
if older adults have more automatic activation than young
adults, then they should have more interference and also
more facilitation. However, if the greater interference with
increasedage is primarily due to decreasedinhibition, then
the amount of facilitation should not systematicallydiffer as
a function of age.
Second,it is important to considerthe manner in which
the inhibition measuresare derived. Interference and facilitation are typically assessedby difference scores,with the
former defined as the time in the incompatible condition
minusthe time in the neutralcondition, and the latterdefined
as the time in the neutral condition minus the time in the
compatiblecondition. Differencescoreshavebeencriticized
becausethe reliability is often low and becausethey have a
high correlationwith the baseline(neutral)score.The question of reliability can be examinedby administeringrepeated
trials in eachcondition to allow the reliability of the difference scoresto be computed. However, Cohen and Cohen
(1983, p. alg suggestthat a bettermethodof taking performanceon one task(neutral)into considerationwhen examining performanceon a secondtask (incompatible)is to partial
variancein the original scorefirst. It is thereforedesirableto
use both the differencescoreand the partialling methodsin
assessingrelationsbetweenage and interference.
Third, it is also possible that what are presumedto be
measuresof specificprocessessuch as inhibition may actually reflect the contribution of a more general or common
influence (cf., Salthouse& Coon, 1994). This possibility
can be examined by partialling out the effects of other
measuresto determine whether the age-relatedeffects are
independentof one another. Measuresfrom severaldifferent
tasks should be included in these types of analysesto
determine if there are effects of a more general factor.
Finally, it is desirableto have multiple measuresof the
relevanttheoreticalconstructs.This is importantnot only to
increase reliability and validity for the reasons mentioned
earlierbut also becauseMacleod (1991) hasnotedthat little
researchhas examined correlations of the interferencemeasuresin different versions of the task.
To summarize, the major goal of this study was to examine age-related effects in measures of inhibition and to
determine the extent to which inhibition may contribute to
the age-relateddifferences in working memory. The working memory construct was assessedwith two tasks, the
reading-spantask and the computation-spantask. Both have
beenusedin severalpreviousstudies(e.g., Salthouse,1991,
1992b. 1992c:.Salthouse& Babcock. l99l: Salthouse&
Coon, 1994; Salthouse,Mitchell, Skovronek, & Babcock,
1989;Salthouse& Skovronek,1992)andhavebeenfound to
yield reliable measures,which have moderatepositive rela-
tions with one another and with several measuresof cognitive performance and moderate negative relations with age.
Interferenceand facilitation were assessedwith three parallel Stroop tasks similar to those used by other researchers
(e.g., see Macleod, 1991, for citations), but specifically
designedto be structurallyparallel (seeFigure l). Subjects
were always instructedto namethe color, the quantity, or the
position of target items, and the stimulus items consistedof
either a neutral string of Xs, a compatible word or number
string, or an incompatibleword or number string. Finally,
additional tests of processing speed were administered to
examine the relation of these measuresto the purportedly
specific measuresof interferenceand facilitation.
Mersoo
Research participants. - Characteristics of the 242
adults who participated in the study are summarizedin Table
l. The participants were recruited from newspaper advertisementsand community organizations.It can be seenthat
the averagelevel of educationwas quite high and that most
participantsreportedthemselvesto be in good to excellent
health.
Tasks.- The different versionsof the Stroop tasks (see
Figure l) were designed to have a similar format. The
stimulusmaterialsin eachconsistedof a pagecontainingtwo
columns of l0 stimuli each. The alphanumericcharacters
were 5 mm high, exceptin the positiontask wherethey were
3 mm high. The rectanglessurroundingthe stimuli were l7
x 30 mm, exceptin the positiontask wherethey were l7 x
37 mm. Researchparticipants were instructed to name the
colors (i.e., red, blue, green, yellow) of the items in the
color version,to namethe quantity (i.e., one, two, three,or
four) of the items in the number version, and to name the
positions (i.e., above, below, right, left) relative to the
internal line in the position version. In all casesthe oral
responseswere to be made as quickly as possibleand were
timed with a stopwatch to a resolution of 0.1 sec. The
participantswere instructedto correct their errors, but be-
of Research
Participants
Characteristics
Tablel. Demographic
50s
60s
70-89
Number
45
32
43
3'7
56
29
9o Female
40.0
68.8
62.8
6'7.6
58.9
51.7
Age
Mean
23.4
35.2
3.1
M.4
2.8
53.7
2.8
65.1
3.1
74.9
4.5
t5.2
2.6
15.5
2.O
16.3
t.6
r5.2
2.4
14.2
2.O
1.6
o.7
2.O
l.l
1.8
0.9
1.9
0.9
2.1
0.9
sD
Education
Mean
sD
Health
Mean
sD
3.3
15.1
1.8
l9
0.8
Nore: Education refers to the number of years of formal education
completed,and healthis a self-ratingon a scalerangingfrom I for Excellent
to 5 for Poor. Becausetherewere very few individualsin the two oldestage
groups,they have beencombinedin the interestof space.
AGING AND INHIBITION
Compatible
Neutral
IncomDatible
Say Color
Say Quantity
Say Position
@
E
fll
r
@
@
@
@
E
I
!
D
Blue
Red
Green
yeltow
lr
E
E
I
BELow
I
l-|;l
|
)oooo(
|
trt t-Fi
E
Ft-r
Figure l. Illustrationofstimulus materialsin the compatible,neutral,and
incompatibleconditionsin the color, number, and position versionsof the
Stroop task.
cause very few errors were uncorrected, the errors were
reflected in longer response times and were not analyzed
separately.
A reading speedtask consistedof a page containing 20
randomly arranged words from the set of four colors, four
positions, and four numbers. As in the Stroop tasks, the
researchparticipantwas instructedto name (read)the items
as quickly as possible, and timing was performed with a
stopwatch.
The tasks were individually administeredto all research
participantsin the following order: reading speed, color,
number, position, position, number, color, and reading
speed.A singlepagewas usedin eachadministrationof the
reading speedconditions, but six pageswere presentedin
each administrationof the other conditions in the order
neutral, compatible, incompatible,incompatible,compatible, and neutral. Four different arrangementsof l0 target
stimuli were preparedfor each Stroop version, and each
arrangementwas presentedonce in each condition. For
purposesof reliability assessment,
the times for the first two
administrationsof eachcondition were averasedand correlated with the averagetimes for the last t*o id.inirtrations
of eachcondition.
The working memory tasks were administered on computers according to the procedures described in Salthouse
andCoon (1994). Briefly, the readingspantask involved the
presentationof a series of sentences,with each sentence
accompaniedby a question with three alternatives. The task
for the researchparticipant was to select the correct answer
to the question and remember the last word in the sentence.
Both the selectionof the correct alternative and the recall of
the target words were entered by keyboard responses.The
individual's readingspanwas the largestnumberof items in
which he or she was correct on both processing(answering
the questions) and storage (recalling the words) in at least
P299
two of three trials. The computation span was very similar
but involved the presentationof a seriei of arithmeiic prob_
lems, with eachproblem accompaniedby threealternatives.
As in the reading span task, the research participant was
instructed to select the correct answer to tlie problem and
rememberthe last item in the problem. Both the selection of
the correct alternative and the recall ofthe tareet disits were
communicated by keyboard responses.Th! indlvidual's
computation span was the largest number of items in which
he or she was correct in both the arithmetic problem and the
recall of the last digit in at least two of three trials. The
readingspanand computationspantaskswere eachadminis_
teredtwice, in immediatesuccession,to allow reliability to
be determined.
The other tasksdesignedto assessprocessingspeedhave
also been describedin previous reports (e.g., Saithouse&
Coon, 1994).The letter comparisonand patierncomparison
tasks required the researchparticipants to make rapi-a
;udg_
mentsabout whethera pair of letter strings (letter compari_
son) or line patterns (pattern comparison) were the same or
different. Stimuli in this task consistedof two columns of
items, and responsesconsistedof writing the letter S (for
-between
same) or D (for different) on the line
the iwo
members of the pair. Two computer-administeredspeed
taskswere the digit digit and digit symbol tests.Thesetasks
were describedby Salthouseand Coon (1994) as follows:
Thedigitsymboltestinvolvedthepresentation
of a codetable
containing
digitspairedwith symbolsand probesof a digit
pairedwith a symbol.Thesubjectwasinstructed
to decideas
rapidlyaspossible
whetherthedigit andsymbolwereassoci_
atedaccording
to thecodetable.If thedigit andsymbolwere
associatedin the code table then the slashkey was to be
pressed,
andifthey werenorpairedin thecodetible thenthe
Z key wasto be pressed.
The digit digit versionof the task
was identicalexceptthat the symbolswere replacedwith
digits,and thusthe yes-nodecisionwas basedon physical
identityratherthanassociational
equivalence.
In bothtasks
subjectswereinstructedto respondasrapidlyandaccurately
(p. l 175).
aspossible
Each of these four tasks (i.e., letter comparison, pattem
comparison,digit digit, and digit symbol) was administered
twice, in immediate succession,to allow reliabilitv to be
computed.
Resulrs
Descriptive characteristicsof the dependentmeasuresare
presentedin Table 2. Reliability was estimatedby boosting
the correlation between the score on the first administration
(averageof two trials for the Stroop tasks) and on the second
administration by the Spearman-Brownformula. It is apparent that most measureshave good reliability, and even the
interference difference score measures have reliabilities
greater than .62. As expected, most of the measures had
moderately large linear relations with age.
Age relations in the Stroop measuresfrom the three tasks
are displayed in Figure 2. It is apparentin Figure 2 that the
quadraticrelations reportedfor a few measuresin Table 2 are
attributable to larger-than-linear slowing among the oldest
participants.
P300
SALTHOUSE AND MEINZ
Table 2. Means, StandardDeviations,EstimatedReliabilities,
and Age Trends for Primary Dependent Variables
Incompatlblo
--o-Ncutral
Proportion
of Variance
Variable
Est. Rel.
Age
Aget
.009
.004
Letter Comparison
Pattern Comparison
9.54 2.69
16.53 4.01
.77
.87
.3t0*
.401*
Digit Digit RT
Digit Symbol RT
739
1 5 1I
.93
.97
.353* .009
.400+ .003
2.50 1.37
4.06 2.1'7
.78
.83
.035+ .001
.018
.000
Reading Span
Computation Span
Reading Speed
Colors ColorsColors Colors Colors -
Neutral
Compatible
Incompatible
Interference
Facilitation
163
415
8.25 1.65
.81
.073*
.000
11.23 2.44
9.63 2.23
19.01 4.82
'7.'78
3.25
1.60 1.60
.89
.84
.253*
.227*
.323*
.217*
.010
.O34*
.010
.002
.005
.020
.009
.005
.Ot2
.oo2
.001
.026*
.013
.O23*
.007
.0r l
.9r
.12
.57
Numbers Numbers Numbers Numbers Numbers -
Neutral
Compatible
Incompatible
Interference
Facilitation
9.82
9.'79
12.32
2.50
0.03
2.21
2.66
2.'72
1.49
t.t4
.92
.62
.41
.229*
.304+
.234+
.031x
. I 33*
Positions Positions Positions Positions Positions-
Neutral
Compatible
Incompatible
Interference
Facilitation
14.05
10.82
t] .53
3.48
3.23
3.30
2.94
4.82
2.34
1.98
.94
.94
.95
.70
.75
-202x
.235*
.193*
.o73+
.001
.vJ
ol
N.)1.,.l€tter Comparisonand PatternComparisonmeasuresare the number
of correctminus the numberof incorrectresponses
in 30 sec,and Digit Digit
RT and Digit Symbol RT are times in msec. Reading Span and Computation
Span measuresrepresentthe largest sequencein which both processingand
recall were correct on at leasttwo of three trials. All remaining measuresare
in sec, with the observed measurescorresponding to the time required to
complete the page of 20 items, and the difference score measures(interference and facilitation) representingdifferencesbetweenobservedtimes.
*/, < .01.
Inspectionof Table 2 reveals that there were relatively
small age-relatedeffectson working memory in this sample.
Correlationsbetweenage and the reading spanand computation spanmeasureswere -. l9 and -. 13, respectively,compared to values of -.37 and -.26, respectively,in another
study(Salthouse,in press)with very similar procedures.The
difference in the two samples is largely attributable to the
older adults in this study performing better than the older
adultsin the other study, which may be relatedto a slightly
greater averageamount of education for the older adults in
this study.
Main effects, and interactionswith age, of education,
health status, and gender for the measuresin Table 2 were
examinedin a seriesof regressionanalyses.Main effectsof
education were significant on many measures(i.e., letter
comparison, pattern comparison, reading span, computation
span, readingspeed,color-neutral,color-compatible,numbers-neutral,numbers-compatible,numbers-incompatible,
positions-neutral, positions-compatible,positions-incompatible, and positions-interference),
but there was a significant interactionof age and educationonly on the positionsfacilitation measure.Health main effects were significant on
numerousvariables(i.e., readingspeed,color-neutral,color-
Compatlblc
....4..'.
l5
Number
8 2 0
I
9 r s
E
c 1
(U
0
o
Position
t5
0
' 2
0
3
0
4
0
5
0
6
0
7
0
ChronologicalAge
Figure 2. Mean time by age decade in the three conditions of the color,
number, and position versions of the Stroop task. Bars adjacent to the data
points representone standarderror.
compatible, numbers-neutral, numbers-compatible,
n u m b e r s - i n c o m p a t i b l e ,p o s i t i o n s - n e u t r a l , p o s i t i o n s compatible, and positions-incompatible), but none of the
interactions with age were significant. Females had significantly higher scores(longer times) on the digit digit, digit
symbol, color-interference, numbers-incompatible,
positions-neutral, positions-compatible, and positionsinterferencemeasures,but there were no significant interactions of age and gender.
Consistent with earlier researchon the Stroop phenomenon (e.g., Macleod, l99l), readingwords (8.25 sec)was
faster then naming colors (11.23 sec), naming quantities
(9.82 sec), or naming positions(14.05 sec). The finding of
smaller facilitation than interference in each task is also
consistentwith earlier research(Macleod, 1991) on these
types of tasks. Although the facilitation effects were small,
AGING AND INHIBITION
t\y *er" significantly greater than zero in the color (r :
15.53)and position(t : 25.31)tasks,but not in the number
task (r : 0.48). Moreover, the patternwas consistentacross
individuals because88.OVoof the participantshad positive
facilitation scoresfor the color task, 5l .iEo for the number
task, and 97.9Vofor the position task. These results, in
combinationwith the moderatereliabilities, suggestthat at
least the facilitation measuresfrom the color and position
tasks were not at a measurementfloor.
Age effects were small to nonexistent on the facilitation
measuresexcept for the number task, suggestingthat age_
relatedeffects on interferencemeasureare not attributablJto
strongerautomaticactivationwith increasedage. The signi_
ficant age effects on the number facilitation -Lorur" appear
due to a shift from positive (benefiQto negative(costf with
increasedage, and thus suggestthat, if anything, there may
be a decreasewith increasedage in the strengthof automatit
activation in this task. Regardlessof how that particular
result is interpreted,however, the lack of an age-related
increasein facilitation implies that the greaterinGrference
with age cannot be attributed to an increasein automatic
activation.
Correlations were computed among the difference scores,
both before and after correction for attenuationby dividing
the correlation by the squareroot of the product of reliabili_
ties.Thesecorrelations,with the disattenuated
correlationsin
parentheses,
were .38 (.57) for the color and numberinterferencemeasures,.52 (.13) for the color and position interfer_
encemeasures,.28 (.42) for the numberandpositioninterfer_
encemeasures,. I 6 (.3 I ) for the color and numberfacilitation
measures,.45 (.69) for the color and position facilitation
measures,and . 14 (.24) for the numberand positionfacilita_
tion measures.It is apparentthat the interferenceand facilitation measuresfrom the color and position tasks have moderate correlations with one another, particularly after
adjustment for reliability, but that the measuresfrom the
number task have smaller relations with other measures.
Although this pattem raisesquestionsabout the extent to
which the measuresall reflect a common construct, a compositeinterferencemeasurewas neverthelessformed by averaging z-scoresfor measuresfrom all three tasks. However,
resultsare also reportedfor each individual measureto allow
the generalizability of the resultsto be examined.
A composite working memory measurewas also created
by averagingthe z-scoresfor the reading span and computation span measures.The correlation between the two measureswas .57, and .7 | after correction for attenuationdue to
unreliability. Composite speed measureswere also created
from the other speed tasks by averaging z-scores from the
similar tasks. The correlations were .67 (.82 after conection
for unreliability) between the letter comparison and pattern
comparisonmeasures,and .84 (.88 after correctionfor unreliability) betweenthe digit digit and digit symbol measures.
Results of the statistical control analyseson the agerelated influenceson working memory aie summarized"in
Table 3 (see Appendix, Note l). All regressionanalyses
except the first were performed hierarchically, with the
increment in R, determinedfor each successivevariable.
The relevant contrastsfor the current purposesare between
the initial age-relatedvariance (Equation l) and the incre-
P30t
ment in variance associatedwith age after control of one or
more other variables. Notice that there was large attenuation
of the age-related effects in working .".iry
when the
measuresof inhibition were controlled, both those from the
differencescoresmeasures(Equations,2,4, and 10, but not
the difference score from the number task, Equation 7) and
thosefrom the sequentialpartialling procedure(Equations 3,
5, 8, and I l). However, the attenuationwas at leastas large
when the measuresfrom the neutral condition ,e." "oi_
trolled (Equations6, 9, and l2), andwhen speedmeasures
from separatetaskswere controlled(Equations13, 14, and
l5). Theseresultsindicate that the phenomenonof attenu_
ated age-relatedvariance in working memory is not simply
attributableto inhibition. Speed, or whatever it is ttrai ii
commonto all thesemeasures,appearsto be more important
than inhibition as a factor contributing to the adult age
differences in working memory.
Becauseof the unusuallysmall relation betweenage and
working memory in this sample,anothersamplewas created
with more typical relationsbetweenage and working mem_
ory. This was achievedby eliminatingdatafrom participants
(n : 9) above the median age (50) with working memory
compositescoresin the top lOVoof the overall distribution,
and data from participants(n : 9) below the median age
with working memory compositescoresin the bottom l0%
of the distribution. The age-relatedvariance (R,) in the
compositeworking memory measurein this sampleof 224
adultswas .104, which is more in line with the valuesfrom
other studiesthan the .033 value in the complete sample.
However, the samepatternof reductionof aqe-relatedvariancein working memory after control of the ipeed variables
was apparent in this more typical sample. That is, the
residual age-relatedvariance in the composite working
memory measure was .035 after control of the color_
interferencemeasure, .099 after control of the numberinterferencemeasure, .050 after control of the positioninterferencemeasure, .06 I after control of the reading speed
measure,.016 after control of the perceptualspeedcomposite measure, and .018 after control of the reaction iime
composite speed measure. The reduction of g4.6%oafter
control of the perceptualspeedcompositeis similar to that
reportedin other studies(Salthouse,1991, 1992c;Salthouse
& Babcock, l99l). For example,a recentstudy (Salthouse,
in press)with the samemeasuresand a similar samplehad a
decreasein the age-relatedvariance in the composite workrng memory measure of 78Vo (from .l4l to .031) after
control ofthe perceptualspeedcomposite,and a decreaseof
797o(from.l4l to .029) after control of the reaction time
speedcomposite(seeAppendix, Note 2).
Relations Among M easures
The next set of analysesexamined the interrelations of the
speedmeasures.The initial analysis determined the effects
of statisticalcontrol of the speedmeasureson the relations
betweenage and the naming measures.The resultsof these
analyses,in terms of the proportion of age-relatedvariance
remaining after control of specified measures,are presented
in Table 4.It can be seenthat there was sisnificantresidual
age-relatedvariance in severalmeasures,birt in all casesthe
residual was only a small fraction of the total age-related
P302
SALTHOUSEANDMEINZ
Table 3. Age-Related Variance in Working Memory MeasuresBefore and After Control of Other Variables
Composite Measure
Reading Span
Computation Span
Controlled Variable
R2
Eq. I
Age
033
.033*
.035
.035+
.018
.018
F4.2
Int. (Composite)
Age
.091
.096
.091*
.005
.079
.086
.079*
.007
.065
.066
.065*
.001
Age
.104
.129
.129
.104*
.025*
.000
.062
.095
.096
.062*
.033*
.001
.104
.t14
.ll8
.104*
.010
.004
Eq.4
C-Int.
Age
.091
.093
.091*
.002
.085
.089
.095+
.004
.059
.060
.059+
.001
Eq. 5
CN
CI
Age
.067
.12
.12
.067*
.045*
.0u)
.038
.091
.o92
.038*
.053+
.001
.070
.091
.093
.070*
.021
.002
Eq. 6
CN
Age
.061
.071
.067*
.004
.038
.M9
.038*
.01I
.070
.070
.070*
.000
Eq.1
N-lnt.
Age
.004
.034
.004
.030+
.001
.036
.001
.035*
.007
.021
.007
.014
Eq. 8
NN
NI
Age
.o9'7
.099
.100
.o9'7*
.002
.001
.063
.063
.069
.063*
.000
.006
.091
.095
.096
*
.091
.004
.001
Eq. 9
NN
Age
.o91
.098
.097*
.001
.063
.069
.063*
.006
.091
.091
.091*
.000
Eq. l0
P-lnt.
Age
.ll3
.t22
.ll3*
.009
.lll
.122
.lll+
.01I
.069
.o73
.069*
.0(M
Eq. ll
PN
PI
Age
.095
.145
.145
.095*
.050+
.000
.054
.120
.t25
.054*
.066*
.005
.098
. tt ]
.ll8
.098*
.019
.001
Eq. 12
PN
Age
.095
.09'7
.095*
.002
.054
.063
.054+
.009
.098
.098
.098*
.000
Eq. 13
Reading Speed
Age
.047
.063
.o47*
.016
.03t
.052
.031*
.o2l
.044
.050
.044x
.006
Eq. 14
PerceptualSpeed
Age
.091
.o92
.091*
.001
.082
.082
.082*
.000
.062
.064
.062*
.002
Eq. 15
RT-Speed
Age
.058
.059
.058*
.001
.05r
.054
.051*
.003
040
040
.040*
.000
Eq. 3
N (Composite)
I (Composite)
Incr. R2
R2
Incr. R2
R2
Incr. R2
Note: lnt. refersto the interferencedifferencescores,CN, NN, and PN refer to the color, number,and positionneutralscores,respectively,and CI, NI, and
Pl ref'erto the color, number, and position incompatiblescores,respectively.
* n ( . 0 1 f o r I n c r e m e n itn R 2 .
AGING AND INHIBITION
Table 4. Age-Related Variance in Naming MeasuresBefore
and After Control of Other Variables
P303
Digit Digit RT
Digit Symbol RT
Criterion Variable
Color
Incompatible
Compatible
Interference
Facilitation
Alone
Neutral
RdSpd
PSpd
Letter Comparison
---€---.
RTSpd
Pattern Comparison
---€---.
.323*
.227*
.21'7*
.010
.038*
.01I
.085*
.021
.201*
.092*
. 1 6 lx
.013
.037*
.Ot4
.033*
.002
.046*
.041*
.028*
.000
Number
Incompatible
Compatible
Interference
Facilitation
.234*
.304*
.031*
.133*
.009*
.018x
x
.031
.098*
.108*
.144*
.o27*
.102*
.007
.030*
.002
.076+
.006
.038*
Position
Incompatible
Compatible
Interference
Facilitation
.193*
.235*
.o73*
.001
.001
.019x
.007
.o42+
.097+
.106*
.o47*
.000
.006
.01l
.001
.000
-o02
.021*
.001
.001
MEAN
.165
.o32
.o92
.018
.019
ReadingSpeed
o
o
(,)
' . . 4
.
Color - Neutral
----
6
Number - Neutral
N
-->-.
Position - Neutral
- - ! -
.n2
.04'l*
Nrre.'Rdspd is ReadingSpeed,PSpdis the PerceptualSpeedcomposite
(lettercomparisonand patterncomparison),and RTSpd is the reactiontime
speedcomposite(Digit Digit RT and Digit Symbol RT).
*/l < .ol.
effects.For example,the residualage-relatedvariancein the
color-interferencemeasurewas only 39Voof the initial agerelated variance after control of the color-neutral measure,
and was only l5.2Voof the initial varianceaftercontrol of the
composite perceptual speed measure. It therefore appears
that most of the age-relatedvariance in these measuresis
sharedwith the other measures.The readingspeedmeasure
is somewhatof an exceptionbecauseit sharesrelativelylittle
variancewith the othernamingmeasures.
Speedmeasuresfrom different tasks were converted to zscores,and plottedby agedecadein Figure 3. Similar trends
are evident for all measures,exceptfor a slightly shallower
agerelation for the reading speedmeasure.This discrepancy
is also apparentin the proportionsofage-relatedvariancein
Table 2 becausethe value for the readingspeedmeasurewas
only .073, comparedto valuesof .202 or greaterfor the other
measures.
The similar age functions apparentin Figure 3 raises the
possibility that one or more common factorsmight contribute to the age-relatedinfluencesin the speedmeasures.This
was examined by determining the proportion of the agerelated variance shared between pairs of speed measures.
These proportions, which are presentedin Table 5, were
calculated by computing the R' for age in the criterion
measure, next computing the increment in R2 for age after
removing the variance in the controlled measure, and then
subtracting the latter from the former and dividing by the
former. To illustrate, the total age-relatedvariance in the
color-incompatiblemeasurewas .323, and the incrementin
variance in that measureassociatedwith age after control of
the color-neutral measurewas .038. The ratio of (.323.038)l .323 : .882, which correspondsto the entry in the 6th
row and 8th column in Table 5.
2 0 3 0 4 0 5 0 6 0
ChronologicalAge
Figure 3. Means by age decade for speed scores expressed in standard
deviation units from the entire sample. Note that the measureshave been
reflected such that higher z-scorescorrespond to faster performance.
The pairs in the table are not symmetric becausethe agerelated variance in the two measuresis not identical. To
illustrate,the age-relatedvariancein the color-neutralmeasure is .253, and the estimateof the proportionof this agerelated variance that was sharedwith the color-incompatible
measureis .984. However, the proportionof the age-related
variance(.323) in the color-incompatiblemeasurethat was
sharedwith the color-neutralmeasurewas .882 (seeAppendix, Note 3).
The meanof all entriesin Table 5 is .863, which indicates
that there is considerablecommonality among the agerelatedvariancein thesemeasures.Substantialcommonality
also exists in the age-relatedinfluences on the derived
measures.This is evident in the proportion of age-related
variance in the interference difference score measuresthat
was sharedwith the other speedmeasures.The proportions
shared with the respective neutral measureswere .608
(Color), 1.00 (Number), and .904 (Position), the proportions shared with the reading speed measure were .258
(Color), .129 (Number),and .356 (Position),thoseshared
with the perceptualspeedcompositewere .848 (Color), .935
(Number), and .986 (Position), and the proportionsshared
with the reactiontime speedcompositewere .871 (Color),
.935 (Number), and .986 (Position). As in the other analyses, therefore, the reading speedmeasureappearsto have a
relatively small proportion of age-relatedvariancein common with the other speedmeasures.
It is also possibleto compute the geometricmean of the
two proportions of sharedage-relatedvariance for a pair of
measures,and then take the squareroot ofthat value to yield
a type of correlation termed the quasi-partial correlation
(Salthouse,1994).The quasi-partialcorrelationis analogous
to a correlation coefficient but, instead of representing the
proportion oftotal variance that is shared, it representsonly
the proportion of age-related variance that is shared. The
quasi-partial correlation for the variance proportions in Table 5 are presentedin Table 6. The mean of the entries in this
table is .886, and the medianis .877.
SALTHOUSEANDMEINZ
P304
Table 5. Proportionof SharedAge-RelatedVariancein SpeedMeasures
Criterion Variable
l0
Controlled Variable
lt
t2
T4
I23-
DDRT
DSRT
LetCom
4
PatCom
RdSpd
5
6 C N
7-CC
8-CI
9-NN
IO-NC
ll-Nl
12 PN
I3 - PC
1 . 1- P l
x
.9n3
.5.58
.652
.266
.5-s0
.473
.694
.5U4
.669
.6'74
.5u6
.567
.5U6
.933
X
.683
.780
.265
.5tt0
.513
.135
.625
.125
.685
.608
.6| 8
.630
.629
.83-5
x
.910
.287
.648
.590
.145
.617
.687
.645
.529
.606
.529
.579
.776
.7-s8
x
.262
.606
.581
.69tt
.628
. 7| 8
.658
-586
.633
.561
.9'73
.986
.932
.999
x
.9'73
.945
.986
.904
.76'7
.986
.999
.973
.959
.735
.846
.767
.8'77
.514
X
.905
.9t14
.893
.949
.870
.826
.838
.119
.705
.828
.167
.907
.595
.952
X
.941
.930
.996
.881
.824
.969
.121
.749
.864
.12t
.833
.378
.882
.762
x
.8ll
.876
.861
.755
.71'7
.777
.830
.939
.852
.943
.620
.939
.926
.974
x
.996
.969
.904
.943
.U9l
.751
.885
.697
.885
.526
.859
.885
.908
.94t
x
.905
.816
.901
.789
.906
.9'70
.803
.951
.s38
.906
.rJ68
991
.962
.996
X
.91-5
.910
. 9t 5
.901
.970
.757
.960
.559
.941
.886
.980
.960
.990
.9'75
x
.9'15
.985
.796
.919
.762
.936
.549
.tatz
.951
.945
.932
.991
.91l
.919
X
.864
.911
.995
.788
.959
.491
.922
.8I 9
.999
.964
.995
.984
.995
.959
X
.ttol
.901
.757
.892
.450
.Ul8
.778
.891
.832
.873
.846
.189
.821
.769
Mean
.603
.645
.640
.619
.952
.tt29
.848
.713
.902
.827
.895
.9ll
.873
.907
.863
Notr,. DDRT - Digit Digit reactiontime: DSRT Digit Symbol reactiontimel LetCom - Letter Comparison:PatCom = PatternComparison;RdSpo :
ReadingSpeed.The remainingvariablesare fiom Strooptaskswith the lirst letterdesignatingthe Stroopversion(Color, Number, or Position)and the second
letterdesignatingthe condition (Neutral, Compatible,or lncompatible).
Table 6. Quasi-PartialCorrelations
Variable
Variable
I 2345-
DDRT
DSRT
LetCom
PatCom
RdSpd
6-CN
7-CC
8*CI
9-NN
IO-NC
ll-Nt
I2 - PN
I3-PC
I4-PI
X
.9'78
.1'7|
.784
.1t2
.'797
.-t59
.850
.836
.843
.883
.853
.821
.858
X
.868
.883
.718
.837
.807
.892
.875
.896
.904
.876
.86'7
.889
X
.9ll
.7l9
.838
.820
.856
.871
.833
.848
.'196
.824
.804
X
.115
.854
.853
.873
.8'77
.892
.890
.867
.8"71
.857
X
.841
.867
.784
.866
.798
.854
.863
.856
.833
X
.963
.965
.951
.950
.942
.940
.925
.919
An exploratoryfactor analysiswas next conductedon the
matrix of quasi-partialcorrelations.The first factor in this
analysiswas associatedwith 89.57oof variance,and inclusion of a second factor accounted for only an additional
3.1Voof the variance.Communality estimatesfor the measures,indicating the proportion of varianceaccountedfor by
the single factor, ranged from .744 (for reading speed)to
.976 (for numbers-interference).
Furthermore,communality
values for the interference measureswere all greater than
.91, indicatingthat a very high proportionof the variancein
thesemeasureswas sharedwith the common factor.
DISCUSSIoN
The results of this study provide some support for the
inhibition construct as measuredfrom similar tasks because
moderate correlations were found among several of the
interference measures after adjusting for unreliability. At
first impression, the results also appear to provide support
X
.921
.962
.967
.934
.925
.982
.877
X
.943
.944
.961
.929
.925
.938
X
.984
.983
.965
.969
.962
X
.974
.949
.9' 73
.940
X
.912
.953
.9'75
X
.9'73
.994
X
.953
for the mediation of the age-related effects on working
memory throughdecreasedinhibition becausecontrol of the
inhibition measuresresultedin a substantialreductionin the
age-relatedvariance in working memory. However, the
interference measures were not independent of the speed
with which many tasks,even sepilrateand distinct tasks,can
be performed. That is, although age-relatedeffects in the
working memory measureswere reducedwhen the variance
in the inhibition measureswas controlled, at least as much
reduction was evident when other speed measures were
controlled.
The first major conclusionof this study is that, contraryto
some interpretations, interference scores from Stroop tasks
are not simply measuresof inhibition because they share
most of their age-related variance with other measuresof
processing speed, even those derived from quite different
types of tasks. The interference measuresmay be a better
reflection of inhibition in samples in which there is little
AGING AND INHIBITION
variation in speed,but the results of this study indicate that
they have relatively little unique age-relatedLffects. As an
example, 84.8Voof the age-relatedvariance in the interference difference score in the color task was shared with the
paper-and-pencilperceptual speedcomposite measure, and
87.l%owas sharedwith the reaction time compositespeed
measure. The residual amounts of age-relatedvariance are
still significantly greater than zero, and thus it can be
inferred that there are distinct age-relatedinfluences on the
interference measuresthat are not shared with other speed
measures.However, it is important to note that the residual
age-related variance is a relatively small fraction of the
amount of age-relatedvariance obtained before taking the
other speed measuresinto consideration. Furthermore, the
attenuation of the age-related effects in working memory
after control of the interferencemeasuresis no greater than
that after control of the other speedmeasures.
The second major conclusion of this study is that the
measuresof speed from the naming tasks, the speed of
making comparisonson paper-and-penciltasks, and the
speed in reaction time tasks appear to share common agerelatedinfluences.The primary evidencein supportof this
conclusion are the high valuesof sharedage-relatedvariance
in Tables 5 and 6, and the high commonalities for the
measuresin a single-factorsolution in the factor analysisof
the quasi-partialcorrelations.
We want to emphasizethat despitethe high proportionof
_
sharedage-relatedvariance,not all ofthe ase-relatedeffects
in these measuresare common, or sharel with the other
measures.Although this is consistentwith the existenceof
specificage-relatedinfluences,the exact contributionof the
specific effects cannot be accurately estimated until the
effects associatedwith the common influenceare first controlled. It is therefore important that both common and
uniqueage-relatedinfluencesbe consideredin assessingany
specific contributionsthat might exist. Moreover, at least
with mostof the measuresin this study, the relativecontribution of the common influence upp"-u.rto be substantially
larger than that associatedwith the specific age-related
influences.
One measurewith a somewhat different pattern than the
others was the speed of reading words. Although the estimated reliability of the reading speedmeasurewas adequate
(i.e., .81), it hadmuch smallerrelationswith asethanmostof
the other speedmeasures(Table 2, Figure 3i and it shared
relatively little age-relatedvariance with the other measures
(Tables 4 and 5). Smaller age relations for reading speed
measuresthan for other naming measureshave also been
reportedby otherresearchers
(e.g., Cohn et al., 1984;Comalli et al., 1962;Houx et al., 1993),andmay be relatedto the
extensiveexperiencemost people have had reading words.
Finally, the implicationof theseresultsfor theoriesbased
on constructssuch as inhibition warrantssome discussion.
Becausethis study only focused on interference measures
from Stroop tasks, it is quite possiblethat alternativemeasures from other types of tasks would provide stronger
supportfor the hypothesizedmediational role of inhibition in
adult age differences in cognition. Of particular interest
would be measuresreflecting different aspectsof the inhibition construct, such as restricting accessof irrelevant infor-
P305
mation into working memory and suppressingprocessing
unrelated to the task goal. Regardlessof the speiific measures that are used, however, the results of this study
indicate that it is important to evaluate the extent to which
the age-relatedeffects in those measuresare independentof
the age-related effects in other measures, such as those
presumed to reflect how quickly simple processing operations can be carriedout.
To summarize, although significant relations have been
reported between age and measureshypothesizedto reflect
inhibition efficiency, and betweenthosemeasuresand measuresreflecting the functioning of working memory, it does
not appear that inhibition mediates the age-related differencesin working memory. At least as assessedby Stroop
interferencetasks, the age-relatedinfluenceson measuresof
inhibition are not independentof the age-relatedinfluences
on other measuresof processing speed. Furthermore, these
other measures account for as much or more of the agerelated variance in measuresof working memory than do ihe
purportedlyspecificmeasuresof inhibition. Until measures
of inhibition are identified with larger independentagerelatedinfluences,therefore,it may be moreparsimoniousto
interpret age-related differences in working memory and
other aspects of cognition in terms of the construct of
processingspeed.
AcKNowLEDcMENTS
This researchwas supportedby NIA Grant R37 96826. We would like to
acknowledge the assistanceof Nate Fristoe, Alice Hardee, and Soo Rhee in
collecting data for this project. We would also like to thank Denise park and
her co-investigatorsfor allowing us to replicate some of the cunent analyses
on their data.
Address correspondenceto Dr. Timothy A. Salthouse, School of psychology, Georgia InstituteofTechnology, Atlanta, GA 30332-0170.
RgpenpNcrs
Cohen, J., & Cohen, P. (1988). Applied multiple regression/correlation
analysisfor the behavioralsciences.Hillsdale, NJ: LawrenceErlbaum
Associates.
Cohn, N.8., Dustman,R.8., & Bradford, D.C. (1984). Age-related
decrements in Stroop color test performance. Journal ofClinical psychology, 40, 1244-1250.
Comalli, P.8., Wapner, S., & Werner, H. (1962). Interferenceeffects of
Stroop color-word test in childhood, adulthood, and aging. Journal of
Genetic Psychology, 100, 47 -53.
Daigneault,S., Braun, C.M.J., & Whitaker, H.A. (1992). Early effectsof
normal aging on perseverative and non-perseverativeprefrontal measures.Developmental N europsychology, 8, 99-114.
Dulaney, C., & Rogers,W.A. (1994). Mechanismsunderlyingreductionin
Stroop interference with practice for young and old adults. Journal of
Experimental Psychology: Learning, Memory and Cognition, 20,
470-484.
Hartley, A.A. (1993). Evidencefor selectivepreservationofspatial selective attentionin old age. Psychologyand Aging, 5,3'7l4jg.
Hartman, M., & Hasher, L. (1991). Aging and suppression:Memory for
previously relevant information. P sychology and Aging, 6, 58j _594.
Hasher,L., &Zacks, R.T. (1988). Working memory, comprehension,and
aging:A review and a new view. In G.H. Bower (Ed.), Thepsychology
of learning and motivation (vol. 22, pp. 193-225). San Diego, CA:
Academic Press.
Houx, P.J., Jolles, J., & Vreeling, F.W. (1993). Stroop interference:
Aging effects assessedwith Stroop Color-Word Test. Experimental
Aging Research,I 9, 209-224.
Macleod, C.M. ( 199l ). Half a centuryof researchon the Stroopeffect:An
integrative review. P sychological Bulletin, I 09, 163-203.
P306
SALTHOUSEANDMEINZ
Panek,P.E., Rush, M.C., & Slade,L.A. (1984). Locus of the Age-Stroop
interference relationship. Journal of Genetic Psychology, 145,
209-216.
Park, D.C., Smith, A.D., Lautenschlager,G., Earles, J., Frieske, D.,
Zwahr, M., & Gaines, C. (1994). Mediators of long+erm memory
performance across the life span. Manuscript submitted for publication.
Rushton,J.P., Brainerd,C.J., & Pressley,M. (1983). Behavioraldevelopment and construct validity: The principle of aggregation. Psychological Bulletin. 94. 18-38.
Salthouse,T. A. ( l99l ). Mediation of adult agedifferencesin cognition by
reductions in working memory and speedofprocessing. Psychological
Science.2. 179-183.
Salthouse,T.A. (1992a). Mechanisms of Age-Cognition Relations in Adult
hood. Htllsdale, NJ: Lawrence Erlbaum Associates.
Salthouse, T.A. (1992b). Working memory mediation of adult age differences in integrative reasoning. Memory & Cognition, 20, 413-423.
Salthouse, T.A. (1992c). Influence of processing speed on adult age
differencesin working memory. Acta Psychologica,79, 155-170.
Salthouse,T.A. (1994). How many causes ,re there of aging-related
decrements in cognitive functioning? Developmental Review, 14,
413-43'7.
Salthouse,T.A. (in press). Influence of processingspeed on adult age
differences in learning. SwissJournal of Psychology.
Salthouse,T.A., & Babcock, R.L. (1991). Decomposingadult age differences in working memory. D evelopmental P sycholo gy, 27, 763-"7'l6.
Salthouse, T.A., & Coon, V.E. (1994). Interpretationof differential
deficits: The caseof aging and mental arithmetic. Journal of Experimental Psychology:Learning, Memory, and Cognition,20, ll'72-1182.
S a l t h o u s eT
, . A . , M i t c h e l l , D . R . D . , S k o v r o n e k ,E . , & B a b c o c k , R . L .
(1989). Effects of adult age and working memory on reasoningand
spatial abilities. Journal of Experimental Psychology: Learning, Memory, and Cognition, I 5 , 5O7-516.
of
Salthouse,T.A., & Skovronek, E. (1992). Within-context assessment
age differences in working memory. Journal of Gerontology: Psychological Scient'es, 47, P l lO-P l2O.
S t o l t z f u s ,E . , H a s h e r ,L . , Z a c k s , R . T . , U l i v i , M . S . , & G o l d s t e i n ,D .
(1993). Investigationsof inhibition and interferencein younger and
older adults. Journal of Gerontology: Psychological Sciences, 48,
Pl79-P188.
ReceivedFebruary 13, 1995
AcceptedJune9,1995
Appendix
1. Because there were significant effects of education on the
working memory measures,analyseswere also conductedto determine the age-relatedvariance in the measuresof working memory
after control of education. The reductions in age-relatedvariance
were relatively small (i.e., from .033 to .029 for the composite
measure,from .035 to .032 for the reading span measure,and from
.018 to .015 for the computation span measure),and thus the
education variable was ignored in subsequentanalyses.
2. Similar results are also evident in a recent study by Park et al.
(1994). In their study, 288 adults between20 and 87 yearsof age
performed a battery of tests including a Stroop Color-Word Test,
the letter comparison and pattern comparison perceptual speed
tests, and the computation spanand reading span working memory
tests. The age-relatedvariance in the composite working-memory
measurewas .128. and it increasedto.l3l after control ofthe
inhibition difference score measure. but decreasedto .009 after
control of the perceptual speedcomposite measure.
3. Similar estimatesof sharedage-relatedvariance were derived
from the data from 80 adults aged2l to 90 years reported by Cohn
et al. (1984). Values computed from the correlation matrix contained in that article indicate that93.5Voofthe age-relatedvariance
in the color-incompatible measure was shared with the colorneutralmeasure,and that 77.37oof the age-relatedvariancein the
color-neutral measure was shared with the color-incompatible
measure.The estimatesfrom the Park et al. (1994) study were also
similar as 9l.5Vo of the age-related variance in the colorincompatible measurewas shared with the color-neutral measure,
and 83 .37oof the age-relatedvariance in the color-neutral measure
was sharedwith that in the color-incomDatiblemeasure.