Joumal oJ Gercntology: PSYCHOLOGICAL
|997. Vol.528. No.5-P216 P228
Copyright 1997 by The Cerontolo!:ical Societ! ofAmerica
SCIENC'S
ControlledandAutomatrcFormsof Memory
andAttention:ProcessPurity andthe Uniqueness
of Age-RelatedInfluences
Timothy A. Salthouse,'Jeffrey P.Toth,rHolly E. Hancock,rand John L. Woodard'
'School of Psychology, Georgia Institute of Technology.
'Department of Neurology, Emory University.
Estimates of controlled and automatic processes hypothesized to underlie performance in a memory task and in an
attention task were derived for 115 participants from 18 to 78 years of age using the process-dissociation procedure.
Participants also performed speed and neuropsychological tests that were suspected to he negatively rclated to age.
Process estimates showed good reliabilfu (from .76 to .98), and the qualintive distinction between processes t'as
supported by the overall pattern of conelations an ong measures. However, only estimated automatic processes exhibited unique variance, as they were either weakly related or unrelated both to performance on the other tests and
to each other. Estimates of the control processes, in contrast, shared considerable variance with measures from
other tests, and there were no unique, or independent, age-related effects on these measures. The results highlight
the need to distinguish between process purity and the uniqueness of age-related influences in accounting for age
differe nc es in cognition.
l\ fOST measuresof cognitive performancecan be asIVlsumed
to reflect a mixture of several theoretical processes.This is a potentialproblem ifone is interestedin assessingthe relationsamong measures,or betweencognitive
measures and other variables, because the observed relations could be attributable to only some of the processesor
to all of them. For example, if Measure I is determined by
theoretical processesA and B, then any relation between
MeasuresI and2 could be due to processA, to processB,
or to both processes.Identifying the specific process(es)responsible for the relation is important, particularly if processesA and B are thought to be differentially affected by
variablesof interest,suchas age.
In comparingrelationsamong measures.it is also important to be sensitive to the issuesof sample size and reliability. Sample size needs to be considered becausethe precision of an estimated relation (i.e., the narrowness of the
confidence interval) is inversely related to the size of the
sample. Reliability also cannot be neglected becauseif the
reliability of a measureis low, then there is little systematic
varianie in that measure available to be associated with
other measures.This is of particular concern with difference scores (such as priming measures)becausethe reliability of such scoresis often lower than that of either of the
constituentscores(e.g.,Cohen & Cohen, 1983,p. 69).
A primary goal in this project was to examine theoretical
process estimates from two different tasks to determine
their relations with each other, with chronological age, and
with a variety of other cognitive measures.The relations
among processestimatesare interesting becauseif some of
the estimates are assumed to reflect a similar theoretical
construct, then one should expect a positive correlation between them. In contrast, little or no correlation would be
P2t6
expectedbetweentheoreticalestimatesif they were postulated to reflect different constructs.
Relations with individual difference variables such as age
are interesting becauseif it is assumedthat most of the relations with age are attributable to one particular process(e.g.,
A and not B), then relations would be expectedbetween age
and estimatesof processA, but little or no relation would be
expected between age and estimatesof process B. Finally,
relations of the process estimates with other measuresof
performance are relevant to the issue of the distinctiveness
of the hypothesizedconstructs.That is, if two processestimates reflect the same theoretical construct, they should
have a similar pattern of relations with other measures;but
if they reflects different constructs,they should exhibit different patternsof relations with other measures.
When the sample consistsof a wide range of ages,and
several different measuresare available from each participant, it is also possible to investigatethe uniquenessof
age-relatedinfluences.That is, many cognitive measures
have been found to be sensitive to age-relatedeffects, but it
is usually unclear whether these effects are independentof
one another.This issue can be important becauseeven if a
measure was thought to provide a pure estimate of a specific theoretical process, the interpretation of the agerelated effects on that measurewould be quite different depending on whether those effects were unique or were
shared with other measures.For example, discovering that
most of the age-related variance in process A was shared
with other cognitive measureswould be consistentwith the
view that age-relatedeffects on processA were simply one
manifestation of a more general shared or common factor
contributing to the age-related influences on many measures.If this were the case, then at least with respect to
PROCESSPURITY AND UNIQUENESS OF AGE EFFECTS
P21l
age-relatedinfluences,there might be nothing special or
For both conditions the primary dependentmeasurewas the
unique about a theoretically pure process estimate if very
probability of respondingwith previouslypresentedwords.
little of its age-relatedinfluence was independentof the
The rationale underlying these two conditions is that
age-relatedinfluenceson other measures.
memory performance reflects the contributions of both conNote that the issue here concernsthe uniquenessor indesciouslycontrolled recollectionand more automatic.involpendenceof age-relatedinfluenceson various measures,and
untary effects of memory. Automatic mnemonic processes
not the purity or validity of the theoretical processesthe
are hypothesizedto increasethe probability of responding
measuresare postulatedto assess.That is, examining agewith old words in both the Inclusion and Exclusion condirelated effects on a measurein the context of age-relatedeftions, whereasrecollection is hypothesizedto increaseor
fects on other measuresis informative as to the uniqueness decreasethis probability dependingon test instructions.By
of age-relatedinfluences,but it is not necessarilyrelevantto
castingtheseconditionsinto equations(essentiallr.a theory
the issueof what the measureactuallyrepresents.
of performance),one can combine performancein the two
When a variety of measuresis availablefrom individuals
conditionsand therebyderive separateestimatesof the conspanninga wide age range,mediationalrelationshipsamong
trolled and automatic processes.Formally, the probability
the measurescan also be investigatedto determinewhich
of using old words in the Inclusion condition is equal to the
measures,or theoreticalprocesses,are plausiblemediators
probability of recollection(C) plus the probabilir\ rhar rhe
of the age relations on other measures.In the present study
old responseswill automaticallycome to mind rA r u,hen
the California VerbalLearningTest (CVLD was usedas the
recollectionis unsuccessful
(l-C). Thus,p(oldllnclusion)=
primary criterion task for these analyses,with the goal of
C + A(l-C). For the Exclusioncondition.in conrrasr.old
determining the extent to which different measuresmight
words should only be used when they come to mind automediateage-relatedeffectson this task. In the past, statistimatically(A) in the absenceof recollection( l-Cr. hecause
cally controlling measuresof processingspeed has been
successfulrecollectionwould act to excludeprer iouslr pre- A( l{t. Combining
found to reduce the age-relatedvariance on a wide range of
sentedwords.Thus,p(oldlExclusion)
cognitive tasks (e.g., Salthouse,1996); hence, speedmeathese equations,mnemonic control (recollectionris estisureswill be examinedtogetherwith estimatespostulatedto
mated as the differencebetweenperformancein thc' Inclureflect pure measuresof different theoreticalprocesses.In
sion and Exclusionconditions(C = Inclusion-Erclusion).
addition,the relative attenuationof the age-relatedvariance
Given an estimateof C, one can algebraicallvderire an esin the theoreticalprocessestimatesand in the speedmeatimate of automaticmnemonic processes[e.g . A = Exclusureswas examinedbeforeand after statisticalcontrol of the
s i o n/ ( l - C ) 1 .
other type of measure.The reasoningwas that the more funThe theory and assumptionsunderlying the processdamentalmeasuresshould producethe greatestattenuation dissociationprocedurehave been discussedextensivelyin
of the age-relatedeffects in the other measures.
the literature(seeJacoby,l99l; Jacobyet al., 1993:Jacoby,
To obtain estimatesof theoretical processesin both a
Begg, & Toth, in press;Reingold & Toth, 1966:Toth, Reinmemory and an attentiontask, we usedthe process-dissoci- gold, & Jacoby, 1994, 1995b); we refer the interested
ation procedure(Jacoby, l99l; Jacoby,Toth, & Yonelinas, readerto thesesources.We should note, however.the cen1993),a generalanalytic techniquedesignedto separatethe
tral assumption underlying the above equations; namely
contribution of controlled and automatic processesto task
that controlled and automaticprocessesmake independent
performance.A stem-completiontask similar to that used
contributionsto performance.Support for this assumption
by Jacoby et al. (1993) was used to assessthe contribution
comesfrom experimentsshowing invariancein one paramof these processesto memory performance. Participants
eter (e.g.,A) acrosslevels of a variable assumedto selecwere presentedwith a list of words under incidentalencodtively affect the processesindexed by the other parameter
ing instructionsand later were presentedwith a seriesof
(C). To date, variablesproducing this pattern in stem- or
word-stems(e.g., tru _ _) correspondingto the previously
fragment-cuedrecall include age (Jacoby,Jennings,& Hay,
presented(old) words as well as unstudied (new) words.
1996),divided attention(Jacobyet al., 1993),and levels-ofAccompanying each test stem was a responseinstruction
processing(Toth et al.,1994). The conversepattern(invari(OLD or NEW), which indicated to participanrswhether
ancein C acrosschangein A) has also been shown (Hay &
they were to include or exclude previously presented Jacoby,1996).
words. For the Inclusion (OLD) condition, participants
In the present study we used the equations described
were to complete each stem with a previously presented above to estimatecontrolled and automaticmnemonic proword or, if they could not remember a studied word, to
cessesfor each researchparticipant.To assessreliability of
complete the stem with the first word that came to mind.
theseestimates,each participantstudiedand was testedon
The Inclusion condition is similar to a traditional cued retwo different setsof (intermixed) words. A novel aspectof
call task, but with instructionsto guesswhen recollection
the present study concerned the testing conditions under
fails. In the Exclusion (NEW) condition, the participants
which mnemonic estimateswere obtained.A potential
were to complete each stem with a word that had not been
problem with the inclusion/exclusioninstructionsis their
presentedbefore; that is, they were to generatenovel comcomplexity (Graf & Komatsu, 1994). To ensure parricipletions for the stems.As in the Inclusion condition, if the
pants' understandingof instructions, most previous proindividual could not remembera previouslypresentedword
cess-dissociation
studieshave used highly trained examin(and thus did not know what word to avoid). he or she was
ers in a one-on-onetestingsession;however,this strateer ir
to complete the stem with the first word that came to mind.
labor intensive and limits the conditions under which rhe
P218
ETAL.
SALTHOUSE
procedure can be employed. As part of a larger effort to increasethe generallty of the procedure (see Hay & Jacoby,
1996;Jacoby,Jennings,& Hay, 1996),we automatedall aspectsof the experimenton computer (i.e., stimuluspresentation, instructions,and responsecollection) and had the
testadministeredby relatively naive examiners.
To estimate controlled and automatic processesin the attentional domain, we used a spatial S-R compatibility task
modeled after Toth et al. (1995a).In this task, lefr (<) or
light-pointing (>) arrows were presentedin the middle, left,
or right portion of a computer screen,and participants were
instructed to make left- and right-handed responsesthat
correspondedto the arrow's direction (e.g., a right-handed
responseto the presentationof '>'). Previous work has
shown that, analogousto Stroop interference,participants
are unable to avoid processingof the irrelevantdimension
(the arrow's spatial location) as shown by faster reaction
times (RTs) when the arrow's direction and spatiallocation
are congruent(e.g.,a right-pointing arrow on the right side
of the screen)as comparedwith when they are incongruent
(e.g.,a righrpointing arrow on the left side of the screen).
Toth et al. (1995a)hypothesizedthat performancein this
task could be describedas reflecting two independentinfluences,controlledprocessingof stimulusform (i.e.,arrow direction)and automaticprocessingof spatiallocation.Similar
to the strategyusedfor cued-recall,they developedequations
for performanceon congruent and incongruenttrials that
were usedto derive separateestimatesof spatial(S) and form
(F) processing.Specifically,they hypothesizedthat the probability of a correct responseon congruent trials is equal to
the probability that spatial processingdirects responding,
plus the probability of respondingon the basisof form when
spatial location does not direct responding;that is, p(correctlcongruent)= S + F(l-S). Correct respondingon incongruent trials, in contrast,should only occur when responding
is based on form and not spatial location; thus p(correctlincongruent) = F(l-S). Subtracting the proportion correct on
incongruent trials from that on congruent trials provides an
estimate of spatial processing(S = congruent-incongruent).
Given S, an estimateof form processingcan be derived algebraically[F = incongruentl(l-S)1.
As with cued-recall,the central assumptionunderlying
the equations for this attention task is that spatial and form
processesmake independent contributions to performance.
Toth et al. (1995a) found supportfor this assumptionwith
the discovery that a manipulation of the proportion of congruent trials influencedspatial,but not form, processingestimates (seealso Lindsay & Jacoby, 1994).Additional evidence for the validity of the estimateswas provided by
showing that form, but not spatial, estimatesaccurately predicted observed performance on arrows presented at fixation (neutral trials) where spatial processingis minimal.
Note that, unlike most attention tasks in which RT is the primary dependentmeasure,Toth et al. (1995a)used a deadline
procedure to obtain proportional data appropriate for use with
the process-dissociation
equations.That is, they required participants to respondwithin 500 ms after stimuluspresentation
to allow assessmentof the accuracy of performance. In the
presentstudy we did not useresponsedeadlinesbecauseof the
well-established relations between age and speed of process-
ing (e.g.,Salthouse,1996).Instead,we usedan RT version of
the task and usedpost hoc deadlinesto analyzethe probability
ofa correctresponseat successivetime intervals.This strategy
allowed us to examine both RT and accuracy as a function of
ageand to reconstructthe time-courseof both observedbehavior (i.e., performanceon congruent,incongruent,and neutral
trials) and processestimates(S and F). Reliability of the process estimateswas assessedby administeringthree separate
blocks of lfi) trials, eachblock containingall trial types.
In order to examine the age relations on the estimatesof
controlledand automaticprocessingin the contextof agerelations on other variables, all participants performed several
additional cognitive tasks that were hypothesizedto be related to age in varying degrees.Severalcomputer-administered and paper-and-pencilspeedtasks were used to assess
highly age-sensitiveperceptualspeedabilities,and two fluency tasks were used to assessperformancein situations
whereboth speedand word knowledgemight be relevant.A
spatial test of line orientationwas administeredbecauseit
was assumedto reflectright hemispherefunctioning,which
has been hypothesizedto decline with increasingage; the
Trail Making Testwas administeredbecauseit is sometimes
postulatedto be sensitiveto frontal lobe functioning,which
has also been hypothesizedto decline with age. Finally, an
episodicmemory task was administeredto provide an agesensitivemeasureof verbal memory that could function as a
criterionmeasurein someof the analvses.
Mersoo
Participants
Of the 124 adultsexamined,completedata were obtained
from l15 between18 and 78 yearsof age.Participantswere
recruitedfrom appealsto groups and acquaintances.The primary criteria for inclusion in the study were that the individual had to be in reasonablygood health, had to have completed at least I I years of education,and was not currently a
student.Descriptivecharacteristicsof the sampleare sununarized in Table l. [Nine participantswere excludedbecauseof
incompletedata.Two participantsin the stem completion task
skipped 25%oor more of the trials, apparently becausethey
kept the ENTER key depressedduring the presentationof the
test items.The datafrom anotherindividual were lost because
the computerwas inadvertentlytumed off in the middle of the
stem completion task. Six participantsin the arrow task used
a reversedstimulus-responsemapping (i.e., they consistently
made left-handedresponsesto right-pointing arrows,and vice
versa), and hence their data were deletedfrom the analyses.
Three other participantshad missing data but were kept in the
overall sample.One had missing data on the digit symbol RT
test, and that value was replaced with an estimatepredicted
from the individual's age and scoreon the digit-digit RT test.
One individual each had missing data on the CVLT recognition measureand the Judgmentof Line OrientationTest, and
in both casesthe mean of the age group was usedas the estimate for the missing value.l
Procedure
To maximize convenience for the participants and to facilitate recruiting, most of the testing was conducted in the
PROCESSPURITYAND UNIQUENESSOF AGE EFFECTS
P2t9
Table l. Descriptive Characteristicsof ResearchParticipants and Summary Measuresof performance
Age Group
N
Age
VaFemale
Education
Health (l = excellent,5 - poor)
Satisfaction
Rating
Limitations
Cardiovascular surgery (7o)
Blood pressuremedications(7o)
Head injury (7o)
Neurological treatment (7o)
Vocabulary (no. correct out of l0)
Synonym
Antonym
CVLT (no. of items correct)
List A
Trial I
Trial2
Trial 3
Trial 4
Trial 5
List B
Immediate free recall
Immediatecued recall
Delayed free recall
Delayedcued recall
Recognition(no. correct)
Arrow
Neutral RT
CongruentRT
lncongruentRT
Form estimate @ 500 ms
Spaceestimate @ 5fi) ms
Stem completion
p(oldlinclusion)
p(oldlexclusion)
p(passlold-incl)
p(passlold-excl)
p(passlnew-incl
)
p(passlnew-excl)
Controlledestimate
Automatic estimate
Reactiontime (ms)
DDRT
DSRT
Paper-and-pencilspeed(no. items/sec)
SD Copy I
SD Copy 2
XO comparison I
XO comparison 2
Letter comparison I
Letter comparison 2
Pattem comparison I
Pattern comparison 2
Fluency (no. items in 60 sec)
VF-C
VF-F
VF-L
VF-animals
VF-furniture
VF-vegetables
Judgment of Line Orientation
Trail Making (no. seconds)
Trail A
Trail B
l 8-39
40-59
60-78
Age Conelation
40
29.0(4.8)
57.5
l s . 5( l . 7 )
38
4 9 . 1( 5 . 1
)
50.0
ts.2(2.s)
69.2(5.1)
51.4
l s.3(2.6)
-rt
2.4(O.s)
2.2(O.8)
1.4(0.6)
0
s . 0( 2 2 . 1 )
s . 0( 2 2 . t )
2 . s( 1 s . 8 )
2.2(O.8)
2.3(0.e)
l.s (0.8)
0
15.8(37.0)
2.6(16.2)
s.3(22.6)
1. 8 ( 1 . 0 )
l 3 . s( 3 4 . 7 )
43.2(s0.2)
2.7(t6.4)
8.1(27.'7)
_.19
-.06
.18
.26*
.41*
-.07
.il
5.s(2.9)
4.9(2.9)
6 . 3( 2 ; 7 )
5 . 2( 3 . t )
8 . 1( 2 . 2 )
6.6(3.2)
.40*
.24*
7. 9 ( 1 . 9 )
to.s (2.4)
t.3 (2.6)
t2.s(2.4)
12.8(2.7\
7.0(2.4)
l 1 . 9( 2 . 8 )
t2.s(2.4)
12.2(2.8)
12.6(2.6)
t 4 . 9( t . 2 )
6.6( 1.6)
e.6(2.3)
to.4(2.s)
l 1 . 3( 2 . 3 )
fi.1 (2.3)
6.s ( r .e)
r 0 . 6( 2 . 8 )
l r . 8( 2 . 5 )
1 0 . 8( 2 . 8 )
t 2 . 0( 2 . s )
r4.0(l.7)
6 . 2( 2 . t )
8.2(2.6)
9.4(2.7)
t0.t (2.1)
1 0 . 7( 2 . 8 )
6 . 4( 1 . 8 )
8.4(3.3)
9.7(3.0)
e.0(3.2)
9.6(2.8)
t3.2(2.t)
-.35*
-.38*
-.344
_.40*
_.40*
-.t]
-.45*
_.41*
-.44*
-.43*
-.41*
434(64)
434(79)
47t ('74)
.688(.259)
.136(.122)
4ee(61)
49s(64)
55l (74)
.403(.26e)
.221(.t20)
600( l 42)
565( l 23)
66t (147)
. l 83 (.236)
.24s(.143)
.59*
.52*
.62*
-.644
.33*
.543(.094)
.234(.r81)
.088(.093)
.r0e(.092)
.22t (.t92)
.098(.099)
.309(.l 94)
.299(.l 83)
.480(.096)
.236(.l s3)
.085(.089)
.t2o (.124)
.2w (.174)
.r0l (.124)
.24s(.188)
.284(.l 50)
.464(.1l3)
. 2 5 5( . 1 r 3 )
. 1 0 8( .r 1 4 )
.ll4(.138)
.232(.226)
. 1 0 3( . 1 5 9 )
.239(.184)
.277(.123)
-.30*
-.M
.o7
.o2
.03
.01
6 6 5( r 0 l )
1 4 2 8( 1 8 6 )
744 (95)
1693(280)
92O(22s)
20'16(4s6)
l.6e(0.3s)
1 . 7 0( 0 . 3 1 )
| . r8 (0.24)
1.22(O.24)
0.38(0.08)
0.33(0.08)
0.64(0.l4)
0.58(0.12)
t.s9 (o.32)
1.62(0.29)
1 . 0 3( 0 . 3 1 )
l.0e(0.32)
0.33(0.09)
0.3r(0.08)
0 . 5 6( 0 . 1 1 )
0.s4(0.09)
l .3l (0.31
)
t.3'7(0.26)
0.87(0.16)
0.90(0.20)
0.27(0.09)
o.23(O.t2)
0.46(0.r0)
0.44(0.10)
16.2(4;7)
14.9(4.2)
1 5 . 0( 4 . r )
20.6(5.1)
12.e(2.8)
13.8(3.3)
12.7(1.7)
14.9(4.9)
13.7(4.6)
(3.3)
1,3.4
18.8(s.7)
12.3(3.4)
14.0(3.9)
12.o(2.2)
14.8(s.2)
14.6(4.7)
l3.s (4.9)
r7.1(4.6)
11 . 8( 3 . 3 )
r4.0(3.3)
r2.1(2.3)
21.0(4.6)
s3.6(20.3)
26.2(7.e)
66.1(21.3\
32.9(10.s)
87.2(36.6\
2.0(0.8)
2 . r( r . 0 )
-. l+
-.0'7
. 5 8*
.65*
-.4'7*
-.49*
-.46*
-.52*
-.41*
-.5 1*
_.47*
-.l5
-.07
_.26*
-. l4
.03
-.l3
. 5 3*
.46+
Notes:CVLI = Califomia Verballrarning Test;RT = reactiontime; DDRT = digirdigit reactiontime; DSRT = digit-symbolreactiontime; VF = verbalfluency
*p < .05.
P220
SALTHOUSE
ETAL.
participants' homes. The testing sessionwas precededby a
short description of the tasks, and written informed consent
was then obtained.The sessionlasted between 1.5 and 2
hours, depending on the individual's pace and desire for
breaks. All participants received the tests in the following
order: background questionnaire, SD copy, XO comparison, letter comparison, pattern comparison, synonym and
antonym vocabulary, digit-digit RT, digit-symbol RT, arrow
RT, stem completion, CVIjI, letter and category verbal fluency, Judgment of Line Orientation, delayed portion CVLT,
and Trail Making Test PartsA and B. The background questionnaire contained a variety of questions about health status and education(Table l).
Speed tests. - The letter comparison and pattern comparison testswere identical to those used in earlier studies
(e.g., Salthouse,1996). These paper-and-penciltests consist of pairs of 3, 6, or 9 letters (letter comparison)or line
segments(pattern comparison),with one-half of the pairs
differing in one element. The task for the participant was
to decide whether the two membersof the pairs were the
same or different, and to write the letter S (for same) or D
(for different) between the members of the pairs as
quickly as possible.Two separateadministrationsof each
test were presentedwith 30 secondsallowed in each administration. For easeof comparisonwith other variables,
the scores were converted to the number of items completed per second.
Two simpler paper-apd-penciltests were designed to assesspresumedcomponentsof the comparisontests.In the
XO comparisontest the items consistedof pairs of letters
that were either both X, both O, or an X and an O. As in the
other comparison tests, the participant was instructed to
classify the pairs as sameor different as quickly as possible
by writing the letter S or the letter D between the two members of the pair. This test was assumedto require a very elementary type of comparison, of a single element, relative to
the letter comparison and pattem comparison tests, which
containedmultiple elements.The SD copy test consistedof
columns of blank lines adjacent to columns of the letters S
and D intermixed. The task for the participant was to copy
the indicated letter as rapidly as possible on the blank line.
This test was assumedto involve some of the same sensory
and motor requirements of the comparison tasks but without any requirement for comparison. Each of these tests
was administered twice with 30 seconds allowed for each
administration. The scores in each test were the number of
items completed per second.
The digirdigit and digit-symbol RT taskshave beenused
in numerousrecentstudies(e.9., Salthouse,1996)and consist of a choice RT responseto a pair of items presentedin
the middle of the computer screen. In the digit-digit task
the items were digits, and in the digit-symbol task the top
item was a digit and the bottom item was a symbol. In both
versions a code table was presentedat the top of the screen,
but in the digit-digit version it merely consistedof identical
digits, whereas in the digirsymbol version it contained
pairs of digits and symbols. The task for the participant was
to decide, as rapidly as possible, whether the pair of items
presented was the same or different, either according to
physical identity (digit digit) or according to correspondence in the code table (digit symbol). Same responses
were indicated by pressing the / key, different responsesby
pressingthe Zkey. A practice block of 18 trials preceded
the test blocks of 27 trials each in both the digirdigit and
digit-symbol versions of the task. Average accuracy was
greater than 95Vo in both of the tasks and was not significantly correlated with RT; performance in each task is
therefore summarized in terms of the median RT in milliseconds(ms).
Vocabulary tests. - The vocabulary tests were identical
to those used in several earlier studies (e.g., Salthouse,
1993), and consistedof l0 five-altemativemultiple-choice
synonym items and l0 five-alternativemultiple-choice
antonym items. Five minutes were allowed for the completion of the two parts of this test. The scoreswere the numbers of words answeredcorrectly in each part of the test.
Anow Task. - Stimuli in the arrow task consistedof
right- (>) and left- (<) pointing arrows presentedalong the
medial-horizontalaxis of the computer screen in one of
three locations:left, right, or center.Trial-type (congruent,
incongruent,neutral) was defined by the relationship between the location and direction of each arrow: For congruent trials, the direction and location of the arrow denoted
the same response(e.9., a left arrow on the left side of
screen);for incongruenttrials, direction and location denoted oppositeresponses(e.9.,a left arrow on the right side
of screen);and for neutral trials, the arrow was presentedin
the center of the screen and had no apparent spatial displacement(relative to the fixation point). In each of three
blocks participantsrespondedto 100 consecutivetrials consisting of 40 congruent, 40 incongruent, and 20 neutral trials. Trial types occurredrandomly within each set of l0 trials so that, although trial type could not be predicted from
trial to trial, the proportion of trial types was consistent
throughouttesting.
Participants were instructed to sit a comfortable distance
from the computer screen. At a distance of 50 cm, each
arrow subtendeda visual angle of approximately l.l"
square and, for congruent and incongruent trials, appeared
approximately 8" to the left and right of fixation. Prior to
testing, participants read instructions and were shown visual examples of the arrows and trial types. Instructions
stressedboth speed and accuracy.The test began immediately after a practice block of 50 trials. Each trial began
with a fixation point (asterisk) presentedin the center of the
screenfor 250 ms. Immediately upon its removal, an arrow
was presentedfor approximately 90 ms. Participantsrespondedby pressing the Z key for left-pointing arrows and
the / key for right-pointing arrows. Responsetiming began
at arrow onset and continued until a response was made.
After a responsewas detected, the screen was cleared and
remained blank for I second, at which time the fixation
point for the next trial was presented.
Stem completion task. - Critical stimuli in the stemcompletion task consisted of 120 five-letter words, mostly
nouns, ranging in frequency from I to 200 (Kucera & Fran-
PROCESSPURITY AND UNIQUENESS OF AGE EFFECTS
cis, 1967). The words were divided into three sets (1-3),
and each set was further divided into two subsets(A and B)
of 20 words each (mean frequencyof the six subsetsfrom
28.8 to 31.9). Two of the sets(1 and 2) were used as studied words. Stemscorrespondingto all of the critical words
were presentedat test under either inclusion (subsetslA,
2A, and 3.A)or exclusion(lB,2P , and 38) instrucrions.
As
is common in individual difference research,all participants received the same items for a given condition to
avoid confoundingexperimentaltreatmentwith the characteristicsof participants.However, the presentationorder of
items at study, and within conditions at test, was randomized for each participant. In addition to the 120 critical
words, therewere 16 buffer words (8 primacy,8 recency),2
of which were usedas examplesin the test instructions(see
below). Word stemswere createdby replacingthe last two
lettersin each word with two underscores.The word stems
were unique within the set of critical words, and each had
at leasttwo completions.
All aspectsof the task were presentedon computersand
were self-paced.Study instructions informed participants
that a list of words would be presentedthat they were to
rate for pleasantness
on a scaleof I to 5 ( I = most unpleasant and 5 = most pleasant).They were told that their ratings
should be basedon the meaning of the words, but that we
were interested in their first impression, so they should
make their ratings quickly. No mention was made of the
subsequentmemory test. Study words remained on the
screenuntil the participant respondedby pressing one of
the number keys (from I to 5) on the keyboard. Once an
appropriatekey was pressed,the next word was automatically presentedafter a blank-screendelay of I second.
Test instructionswere presentedimmediately following
study.Participantswere informed that their memory would
be testedfor words presentedin the pleasantness-rating
task
using word-stems["the first three lettersof five-letterwords
followed by two dashes(e.g., pea_ _)"1. They were also
told that, in addition to word stems,the messageOLD or
NEW would appearaboveeach stem,and that they were to
responddifferently dependingon this message:
If themessage
is OLD yourjob is to completethestemwith
a word presentedin the pleasantness
task; that is, try to
completethestemwith an old word.If themessage
is NEW
yourjob is to completethe stemwith a wordthatwasNOT
presented
in the pleasantness
task;that is, try to comeup
with a new word. Regardless
(OLD or
of the message
NEW), if you cannotremembera word from the pleasantnesstaskthatfits into the stem,thenjust completethe stem
with the first live-letterword thatcomesto mind.All of the
stemscanbe completedwith morethanoneword so try to
comeup with a completionfor all of them.However,do not
usepluralsor propernouns.Also,ifthe message
tellsyou to
give a NEW word,but all you canthink of is ONE completion that you are SUREis old, thenit is okay to passthat
stemby enteringxx.
Participantsmade their responsesby typing in two letters,
which appearedon the screen under the stem's two underscores,and then pressing the ENTER key.
Following test instructions,the computerled participants
through four practice test trials. The first two used stems
P22l
correspondingto the last two buffer words in the study list;
the last two practice trials used new stems.The first and
third practicetrials were Inclusion (OLD) trials, the second
and fourth were Exclusion (NEW) trials. For each practice
trial, the computer representedthe appropriate instructions
and provided feedback concerning the participant's responses.For example,if a participantrespondedwith a plural word, the computer informed him or her of the error and
presentedthe test stem again until an appropriate response
was provided.Following practice,the participantwas given
the option of beginning the actual test or repeating the practice phase. No feedback was provided during the test
proper.
Test stemswere presentedin the centerof the screen,in
large lower-caseletters,and in white againsta black background. The Inclusion (OLD) and Exclusion (NEW) cues
were bright green and bright red, respectively.They appeared500 ms prior to the presentationof the test stem and
were positionedin the center of the screenjust above the
stem.Stemsand responsecuesremainedon the screenuntil
participantsresponded;if l5 secondselapsedwithout a response,the computerbeepedand the message"Pleaseenter
a response"appearedbelow the stem.Once a responsewas
entered,the responsecue for the next trial (OLD or NEW)
appearedimmediately.
Other tests. - The CVLT (Delis, Kramer, Kaplan, &
Ober, 1987) was administeredaccordingto the instructions
in the manual.There are two lists of 16 words each in this
test, with fbur words from each list in each of four categories (i.e., spices,fruits, clothing, tools). The procedure
involves the presentationof five immediatefree recall trials
with List A, one immediatefree recall trial with List B, recall of List A, and cued (by category)recall of List A. After
a 20-minutedelay occupiedby other activities,there was a
delayedfree recall test of List A, a delayedcued recall of
List A, and a delayedrecognitiontest for items from List A.
Two different types of verbal fluency tests were administered. In the letter fluency tests the participants were allowed I minute each to say as many words as possiblethat
began with the lettersC, F, and L, with the constraintthat
none of the words should be proper nouns. In the category
fluency test they were allowed I minute each to say as
many words as possiblethat were membersof the animals,
fumiture, or vegetablescategories.The score in each test
was the number of different words produced that were in
the appropriatecategory.
The Judgment of Line Orientation Test (Benton, Hamsher,Varney,& Spreen,1983) was administeredaccording
to the published instructions except that only the odd-numbered items were presented.This test consists of a display
of line segments in different orientations with the participant instructed to select the lines of matching orientations
from a semicircle of numbered lines. The score is the number of test lines matched correctly. This abbreviated form
has been demonstrated to have acceptable reliability and
validity (Woodard et al., 1997).
The Trail Making Test (Reitan, 1992) was also administered according to the published instructions. This test consistsof two versions,A and B, and in both versionsthe test
P222
SALTHOUSE
ETAL.
form consistsof ahaphazardly arrangedset of 25 circles on
a piece ofpaper that are to be connectedas rapidly as possible. In the A version of the test the circles containthe numbers from I to 25, and the circles are to be connectedin numerical sequence,but in the B version the circles contain
alternating letters and numbers; thus, the participant has to
connect the I circle with the A circle, the A circle to the 2
circle, the 2 circle to the B circle, etc. Becauseerrors were
very infrequent, the measure of performance in these tests
is the time in secondsto completethe sequence.
Rpsulrs
CVLT analyses.- There are many possible measures
that can be derived in the CVLT, but most have similar relations to age (Table l) and are not independentof one another. A principal componentsanalysiswas thereforeconducted to identify a parsimoniousset of variablesfor later
analyses.Two componentswere identifiedwith eigenvalues
greater than l, the first accounting for 68.3c/oof the variance, and the secondaccounting for 9.8%oof the variance.
All of the variablesexcept List B recall had moderateto
high loadings(between.62 and .92) on the first component,
and List B recall was the only variable with a high loading
(>.5) on the secondcomponent.Further analysesrevealed
that the sum of the scoresacrossthe first five trials in List A
was correlated.96 with the first principal componentscore
and only .15 with the second component score, whereas
List B recall was correlated.74 with the secondcomponent
score and only .49 with the first component score. These
two variables were therefore used to summarize CVLT performancein all subsequentanalyses.
Arrow analyseJ.- Two separatesets of measureswere
obtained in the arrow task. One set consistedof the median
reaction times (RT) on neutral trials (arrows in the middle of
the screen),congruenttrials (e.g.,left-pointingarrow on the
left side of the screen),and incongruenttrials (e.g., leftpointing arrow on the right side of the screen). Means of
these medians for each age group are presentedin Table l,
where it can be seen that RTs in each condition increased
with age. Hierarchical regression analyseswere conducted
to determine whether there was a significant influence of
age on congruent or incongruent RT after control of neutraltrial RT. In neither case was the residual age-relatedvariance significantly different from zero (i.e., incongruent,AR2
= .005; congruent,AR' = .002), indicating that there was no
relation between age and either measures of interference
(i.e., incongruentrelativeto neutral)or facilitation(i.e.,congruent relative to neutral) after overall speedwas taken into
consideration.
The secondset of measuresfrom the arrow task was estimates of the probabilities of spatial- and form-based responding. The initial step in these analysesconsistedof
partitioning the data according to responsetime, and then
computing the cumulative probability of a responsein the
neutral, congruent, and incongruent conditions for each
100-ms interval from 200 to 1000 ms. The results of these
computations are displayed in Figure l. Notice that, as expected, the probabilities increase with greater time, and
o
o
c
o
o.
0.9
Young
0.8
an o.7
o
0.6
E
E 0.5
o.4
j
!
Neutral
+
g
-o
G'
.o
I
IL
Conoruent
--%-.
Incongruent
"'u"'
0.3
o.2
0.1
200 300 400 500 600 700 800 9001,000
Middle
o
0.9
o
0.8
o
o
0.7
G
0.6
o
0.5
Neutral
:=
o.4
Conoruent
--:a--
at,
c
CL
.a.""'o""'.El
F"
-o
(t,
0.3
.ct
o o.2
o- 0.'l
Incongruent
"'u"'
200 300 400 s00 600 700 800 9001,000
o
o
c
0.9
ord
o
o.
o
0.8
G
0.6
o
a
0.5
0.4
Neutral
I
0.3
Conoruent
- -.4 -.
o.2
lncongruent
o
-o
I
o-
. 8 "
o.7
0.1
0-
200 300 400 500 600 700 800 9001,000
ReactionTime(ms)
Figure l. Probability of a responsein the neutral, congruent, and incongruent conditions in the arrow task as a function of age group (young l8 to 39, middle = 40 to 59, and old = 60 to 78) and post hoc resoonse
deadline.
PROCESS
PURITYAND UNIOUENESS
OF AGE EFFECTS
there are pronounced age differences such that the functions are shifted to the right with increasing age. It can also
be seen that the functions for the neutral and congruent
conditions were very similar, but that the functions for the
incongruent conditions were shifted to the right relative to
the other two conditions.
The data summarized in Figure I were used to compute
estimatesof the probability of spatial (S) and form (F) processingat each successiveresponsetime interval from the
equationspresentedearlier.The resulting estimatesare illustratedin Figure 2. Notice that F increasedmonotonically
over time, and that there is a rise followed by a decreasein
S. There is a large decrease(i.e., shift of the functions toward longer times) with age in F, but an increasewith age
in S, particularly in the range between500 to 700 ms. Becausethe 500 ms values appearto representthe middle of
the range for all age groups,those values were used in the
subsequentanalyses.(Similar patternsof relationsto other
variableswere apparentwith estimatesderived from deadlines ranging from 400 to 800 ms). As in the data reported
by Toth et al. (1995a), the theoreticaldistinction between
the two processestimateswas supportedby their relations
with observed performance on neutral (center-arrow) trials
that provide a relatively pure measureof form processing:
Correlation coefficients predicting neutral-trial performance
at 500 ms were .96 for F and -. l0 for S.
To assessreliability, estimatesof spatial and form processingat the 500-ms deadlinewere computedfor each of
the three blocks of trials. Using the formula n(avg. r)/
(l+([n-l]avg. r)), the resultingreliability estimateswere
.98 for F (the controlledparameter),and .83 for S (the automatic parameter).Estimatescomputed separatelyfbr each
age group were: F, )oung = .96, middle = .96, and old =
.97; S, young= .80,middle= .7J, andold = .84.
Both parameters from the arrow task had moderate to
strong correlationswith age. The correlation with F was
,"
.
E
F
..a
o'7
Young - Fom
-----l--
0.6
Middls- Fm
----a---
o
t
E
g
o
o-
ou
Old - Fom
..o .
Y@ng - Spa€
0.4
l/ilddle^-Spae
---r'-\---
0.3
Old - Space
' . . . . o ..
0.1
-=R:=
AD
300
,lO0
50o
mO
7OO
800
900
1,000
Deadline (ms)
Figure 2. Estimated proportion of trials based on form or spatial location as a function of age group (young = 18 to 39, middle = 40 to 59, and
old = 60 to 78) and post hoc responsedeadline.
P223
-.64, indicating that increasedage was associatedwith a
lower probability of making a form response prior to the
500-ms deadline. The same pattern was evident at the 400and 600-ms deadlines as the age correlations with these estimates were -.58 and -.53, respectively.The correlation
betweenage and S at the 500-msdeadlinewas .33, indicating that increasedage was associatedwith a higher probability of making a responsebased on the spatial location of
the arrow. A similar relation (r = .44\ was evident at the
600-ms deadline, but the relation was reversed at 400 ms
with the correlationbetweenage and S being -.28. This reversal reflects the different time coursesof the spatial estimatesfor the three age groups apparentin Figure 2.
The resultsjust describedsuggestthat the relative influenceand time-courseof form and spatialprocessesis qualitatively similar in the three age groups. More quantitative
comparisons,however, are complicatedby the increasing
responsetimes as a function of age.To determinewhether
aging had any process-specificeffects on form and spatial
processing,we performed another analysis in which the
functions summarized in Figure 2 were adjusted on the
basis of RTs to neutral (center-arrow) trials. For this analysis we computed an RT for each participant at the point
where he or she was making 5OVocorTectneutral-trial responses.This RT was then usedas the individual's post hoc
deadlinefor assessingprobability corect on congruentand
incongruenttrials. Theseprobabilities,in tum, were usedto
compute relative RT estimatesof F and S for each participant using the same equationsdescribedabove. With this
method of computing estimates,the correlation between
age and F was substantiallyreduced(.05), but the relation
between age and S remained well above zero (.27). This
pattern suggeststhat age differences in controlled (form)
processing,but not in automatic (spatial) processing,are
eliminated when overall responsespeedis taken into consideration(seealso Spieler,Balota, & Faust, 1996).
Stgm completion analyses.- Two observedprobabilities
were of particular interest in the stem-completion task; the
probability of responding with an old item and the probability of not producing a response(i.e., "passing").Age relations for each of these measuresare summarized in Table
1. Notice that there was a significantnegativeage relation
(r = -.30) for the proportion of old items produced in the
Inclusion condition, but there was no significant age relation (r = -.04) for the proportionof old responsesin the Exclusion condition. There were also no significant relations
between age and the probability of passing in any of the
conditions.
Estimates of mnemonic control and automaticity were
computed from the formulas described earlier. The means
of these estimatesfor the three age groups are presentedin
Table l. Although the mean for the controlled processing
estimate was almost 237olower in the oldest age group than
in the youngest, the age correlation was only -. 14. This
weak age relation was surprising in light of previous studies (e.g., Jacoby et al., 1996), and therefore possible reasons for the lack of a stronger age relation were explored.
Becausevocabulary was positively correlated both with age
(r = .34) and with the controlled parameter (r = .32), it is
I
I
I
P224
SALTHOUSE
ETAL.
possible that the current sample was biased in favor of
high-functioning older adults. Indeed, after controlling for
the level of vocabulary, the correlation between age and the
estimateof controlled processingchangedfrom -.14 (n.s.)
to -.27 (p < .01). Although this finding is interesring,without additional information it is not possible to determine
whether the apparent discrepancy between the present
study and earlier studies is attributable to our use of a more
select sample of older adults.
To assessreliability, controlled and automatic processing
parameterswere computed for each of the two sets of items
presentedin the stem task. Correlations were calculated between the two sets of parameters, and these correlations
were boosted by the Spearman-Brown formula to estimate
reliability of the averagesacrossthe two blocks. The resulting reliability estimateswere .76 for the controlled parameter and .82 for the automatic parameter. Estimates computed separatelyfor the three age groups were: Controlled,
young - .84, middle =.67, and old = .76; Automatic, young
= .88, middle = .87, and old = .66.
(r = .09). Furthermore, correlations between the controlled
estimate in one task and the automatic estimate in the other
task were low and not significantly different from zero (i.e.,
r = -.16 and r = -.ll). This patternof resultssuggeststhat
there is some corrunonality in the two measures of controlled processing,but little overlap in what is being assessedin the two measuresof automaticprocessing.
The relatively large negative correlation (r = -.71) between controlled and automatic estimates in the stem task
was surprising, especially given the assumption that these
two processesmake independent contributions to performance (see also Curran & Hintzman, 1995). However, it is
important to emphasizethat although the existenceof this relation indicates that the two parametersare not independent
at the level of individuals, it does not necessarily indicate
that the two processeswere not independentwithin a given
individual. That is, only if correlations could be computed
within an individual would their magnitudes be relevant to
claims about the independenceof the two processes(see
Cowan, 1996;Jacobyet al., 1997).Severalpossibleexplanations of the negative correlation were explored, but none
proved viable. For example, the correlation might have been
atkibutable to the relation of each parameter with a third
variable such as vocabulary,but statisticalcontrol of vocabulary had little effect on the correlation (i.e., changing from r
= -.71 to r = -.68). Moreover, the correlation between the
probability of completing the stemswith old items in the Exclusion and Inclusion conditionswas very low (r = -.06),
and thus it does not appear that the participants simply ignored the instructions and merely tried to recall as many old
words as possibleregardlessof the condition.
As expected, the correlations with other measureswere
hieher for the controlled estimates than for the automatic
Correlational analyses. - Table 2 contains the correlation matrix for all of the major variables in the study including age. Values along the diagonal in parenthesesare
estimatesof the reliability of the variable.
If there are separate constructs representing controlled
and automatic processes,the two measuresof the same
construct should be moderately conelated with one another
but either weakly or not correlated with measures of the
other construct. Inspection of Table 2 reveals that there was
a small positive correlation (r = .26) between the controlled
estimates from the stem and arrow tasks, but little or no
correlationbetweenthe two automaticprocessingestimates
Table 2. Correlations of Maior Variables
r
I
I
I Age
2 Stem-controlled
3 Stem-automatic
4 Arrow-controlled (F)
5 Arrow-automatic (S)
6 Vocabulary
7 DDRT
8 DSRT
9 SD copy
l0 XO comparison
I | [rtter comparison
12 Pattem comparison
13 ktter fluency
14 Category fluency
l5 JudgmentofLine Orientation
l6 Trails A
17 Trail B
t8 cvLn-5
19CVLIB
Mean
SD
2
3
4
5
6'7
I
9
l0
n
t2
X
-.t4 (;76)
- . u - ; 7 1 (.82)
-.64 .26 -.11 (.98)
. 3 3 - . 1 6 .09 -.17 (.83)
.34 .32 -.33 -.07 .10 (.87)
.58 -.22 .03 -.51 .0'7 .06 x
.65 -.24 .07 -.56 .01 .14 .72 X
-.4"1 .20 - . 1 8 . M . 0 1 -.10 -.51 -.s2 (.9s)
-.49 .27 -.12 .53 -.03 -.00 -.52 -.58 .63 (.e2)
-.52 .22 -.04 .53 -.21 .02 -.46 -.49 .31 .43 (.72)
-.54 .30 - . 1 9 . 5 3 - . 1 3 -.02 -.50 -.55 .43 .47 .s2 (.74\
-.14 .26 - . 1 8 . 3 4 . 0 0 .32 -.35 -.29 .29 .37 .21 .31
-.18 .22 - . 1 5 . 3 0 . 1 2 .31 -.33 -.31 .28 .26 .23 .24
- . 1 3 . 1 9 -.14 .41 -.10 .21 -.t9 -.16 -.02 .10 .25 .25
.53 -.24 .04 -.53 .13 .12 .44 .6t -.53 -.55 -.36 -.48
.46 -.25 .06 -.54 .17 -.00 .49 .55 -.27 -.44 -.45 -.48
-.43 .31 -.13 .41 -.08 .16 -.39 -.50 .34 .38 .33 .42
-.17 .1'7 - . 1 4 . 2 1 - . 1 1 .w -.20 -.21 .16 . 1 9 . l t
.22
48.6 0.27 0.29 0.43 0.20 6.0'1 773 1124 1.55 1.02 0.31 0.54
t 7. 3 0 . 1 9 0 . 1 50 . 3 30 . 1 4 2.90 184 417 0.33 0.2'70.09 0.r2
13
t4
15
(.85)
.49 (.66)
.16 .09 x
-.32 -.20 -.19
-.29 -.1'7 -.36
.46 .N
.01
.32 .23 -.08
14.6 15.0 12.3
4.0 3.1 2.1
16
X
.50
-.38
-.13
26.5
9.3
t7
18
19
x
-.43 (.96)
-.24 .49 X
68.7 49.8 6.6
3 0 . 1 1 1 . 1 2.0
Notes: Correlations with an absolute value greater than .24 are significantly (p < .01) different from zero. Values in parenthesesalong the diagonal are
estimatesof reliability. No estimate of reliability was available for the variables with an X in the diagonal. DDRT = digit-digit reaction time; CVLT = California Verbal Learnine Test.
PROCESSPURITY AND UNIQUENESS OF AGE EFFECTS
estimates.For example,the correlationswith the total number of items recalledacrossthe first five tnals of the CVLT
were.3l and.4l for the stem and arrow controlledestimates,but only -. 13 and -.08 for the stem and arrow automatic estimates.It should be noted that this different pattern of relationsis not an artifact of low reliability, because
the reliability estimatesfor the theoreticallyderived parameterswere all between.76 and .98.
Commonfactor analyses.- Many performancemeasures
have moderaterelations with age,and thereforeit is useful to
examine the independence,or uniqueness,of the age-related
influenceson eachof the measures.
One way to assessthis is
by a single common factor analysis(e.g., Kliegl & Mayr,
1992;Salthouse,1994).This procedureconsistsofdetermining what sharedage-relatedinfluencescontribute to the agerelated effects on each measure,and then identifying the
measureswith unique (additional) age-relatedinfluences.
The following stepswere carriedout in our implementation
o[ the singlecommonfactorprocedure.First.measures
expressedin units of time (i.e.,digit-digit and digirsymbol RT
and Trail Making PartsA and B) were converted to reciprocals such that higher scoresrepresentedbetter performance.
Second,all measureswere postulatedto be related to the
common factor, and the common factor was postulatedto be
relatedto age. Measureswith no significantloading on the
common factor had their measure-commonpaths deleted
tiom the model, and the remaining measure-commonand
age-commonpath coefficients were freely estimated.These
parameterswere then fixed to the estimatedvalues when determiningwhetherany of the measureshad significantdirect
or independentrelationswith age.(Note that overall fit statistics are not particularlyinformative in this type of analysis
becausethe focushereis exclusivelyon the age-relatedinlluenceson the variables,and patternsof relationsamong variablesare ignored.)
The results of this analytical procedureare summarized
in Figure 3. The arrows indicatethe relationsthat were significantly different from zero, and the numbersconespond
to standardizedregression coefficients. The moderate to
strongloadingsof most measureson the common factor indicate that the measuressharesubstantialamountsof variance with one another,and the strong negativerelation between age and the common factor indicatesthat increasing
age is associatedwith lower levels of what the measures
have in common. Of particularinterestare the relationsinvolving the estimatesof controlled and automaticprocessing. Notice that the two control estimatesboth loaded on
the common factor but had no independent age-relatedeffects, indicating that all of the age-relatedinfluences on
these measureswere sharedwith other measures.In contrast, the automatic estimate from the stem task did not
have a significant loading on the common factor, and the
automatic estimatefrom the arrow task had a small but negative loading. This pattern suggeststhat the automaticestimates are measuring something that is distinct from that
being measured by the other variables. However, because
the two automatic estimates were not correlated with each
other (i.e., r = .09), they are apparently not measuring the
sametypeof automaticprocessing.
AGE
"7e
P225
-
coMMoN
----*
.36
Anow - Controlled
---->
.78
Anow -AutomEtlc
->
-.21
DDRT
---->
.80
DSRT
---->
.83
SamrDlff
----.>
.64
XO Comp.
-->
.zg
Lettercomp.
--->
.65
Pattemcomp.
----D
.70
LetterFluency
---->
.48
CategoryFluency
----D
.98
JOLO
----D
.27
TnllsA
----D
.67
Tralls B
----*
.73
CVLTA1€
---->
.59
cvLT 81
----D
.29
Stem- Controlled
Stem - Automadc
.21 ---:D
Figure 3. Resultsof the single common factor analysisindicatingdirect
and indirect age-relatedinfluenceson the cognitive variables.DDRT =
digit-digit reaction time; DSRT = digit-symbol reaction time; JOLO =
Judgment of Line Orientation; CVLT = California Verbal Learnine Test.
The letter fluency variable had a unique age-related effect in the opposite direction of the age-common relation.
This may be attributable to the fact that letter fluency is
positively related to level of vocabulary,which is greater
with increasedage in this sample(seebelow).
Mediational analyses.- Table 3 contains the results of
the prediction of CVLT performance from the speed measures and the estimatesof controlled processing.Entries in
the first row correspondto the proportion of variance associated with age and the standardizedregressioncoefficient for
age when it was the only predictor in the regressionequation.
Subsequentrows contain corresponding values obtained
from hierarchical regressionanalyseswhere the variables in
the first column were controlled before examining the age re-
P226
SALTHOUSE
ETAL.
lations on the CVLT measure.It can be seenthat there was a
very large reduction of the age-relatedvariance after control
of the speedvariables.To illustrate, the reduction was 78Vo
after control of the perceptual speed composite (i.e., letter
comparison and pattem comparison), and 9lvo after control
of the digifsymbol reactiontime measure.In contrast,the reduction was only l87o after control of the controlled recolfection parameterfrom the stem completion task, and 75Vo
after control of the form parameterfrom the arrow task.
Finally, Table 4 containsthe resultsof regressionanalyses predicting the age-related variance in the speed variablesand the controlled-processing
parametersafter control
of the other variables.Notice that there was a large asymmetry with the controlled (recollection) parameterfrom the
stem task because statistical control of that parameter reducedthe age-relatedvariancein the perceptualspeedcomposite by only llvo (from .376 to .333), but control of rhe
perceptualspeed composite reduced the age-relatedvariance in the recollection parameterby 78Vo(from .018 to
.004). A similar asymmetry was apparentwith the digitsymbol reactiontime measure;age-relatedvariancein that
measureafter statisticalcontrol of the recollectionparameter was reducedby only 8Vo(from .424 1o.390),but control
of digirsymbol reaction time reducedage-relatedvariance
in estimatedrecollectionby 94Vo(from .0l 8 to .001).
There was less asymmetry in reductionsof age-related
variance with the controlled (form) parameter from the
arrow task, perhapsbecauseit is closely related to speed.
To illustrate,control of the perceptualspeedcompositereduced the age-related variance in the fbrm parameter by
TlVa (from.416 to .119),whereascontrol of the fbrm parameter reduced the age-related variance in the perceptual
speedcompositeby 78Vo(from .316 to.084). Control of the
digit-symbol RT measurereducedthe age-relatedvariance
in the form parameterby 67Vo(from .416 to .136), while
control of the form parameterreduced the age-relatedvariance in the digit-symbol RT measureby 66Vo(from .424
to .144).
Results from several other multiple regressionanalyses
are also of interest. For example, although the age-related
variance in the digit-symbol RT measure was reduced by
SlVo after control of the digit-digit RT measure,the residual
age-relatedvariance was still significantly greater than zero
(i.e.,AR' = .082).Statisticalcontrol of the SD copy measure
reduced the age-relatedvariance in the XO comparison
measureby 92Vo(i.e., AR' = .089, n.s.), and statisticalcontrol of the XO comparison measure reduced, but did not
eliminate, the age-relatedvariance in the letter comparison
(i.e., lSVo reduction, AR' = .096) and pattern comparison
(i.e.,21Voreduction,M' = .145) measures.
Statisticalcontrol of the Trails A measurereducedthe age-relatedvariance
in the Trails B measureby 75Vo,but the residual age-related
variance(i.e.,M' = .055) was still significantlygreaterthan
zero. Finally, statistical control of the vocabulary measure
increasedthe magnitude of the age relations on the two fluency measures,as the correlation with age changedfrom
-.14 (nonsignificant)
to -.28 (p <.01) with the letterfluency
measure,and it changedfrom -. l8 (nonsignificant)to -.32
(p < .01) for the categoryfluencymeasure.
DrscussroN
The presentstudy producedfour new findings of primary
interest.First, processestimatesderived from the process-
Table 3. Predictionof Age-RelatedVariance
in CVLT I-5 Performance
o/o
Reduction
Age alone
Age afier:
LCIPC composite
SD
XO
SD, XO
SD, XO, LC/PC composite
DDRT
DSRT
Stem-controlled
LCIPC composite,stem-controlled
DSRT, stem-controlled
Arrow-controlled(Form)
LC/PC composite,arrow-controlled
(Form)
DSRI anow-controlled (Form)
I
.l8t*
.040
.093*
.074+
.063*
.021
.061*
.011
.149
.045
. 0 18
.04-s
71.9
48.6
I
-59.
65.2
88.4
66.3
90.6
l 7. 7
'15.1
90.1
75.1
-.253
-.344
-.313
-.297
-.t91
-.304
-. 169
-.391
-.270
-.t17
-.2'76
.0t8
.0(X
90.1
97.8
- . 18 7
-.098
Note: The LC/PC composite is the averageof the z-scoresfrom the le!
ter comparisonand patterncomparisonmeasures.DDRT = digit-digit reaction time; DSRT = digit-symbol reactiontime.
Table 4. Prediction of Ase-Related Variance
in Speedand Contr6lledEstimates
Arrow
LC/PC Comparison
R'
Age alone
Age after:
LCIPC comparison
DSRT
Stem
Arrow
.018
- . 13 6
.004
.001
.082
.03s
.o02
.055
.416*
-.645
.ll9*
. 13 6 x
.378*
-.435
-.486
-.621
DSRT
R'
.376x
*.614
.088*
.333E
.084*
-.389
-.583
-.378
.424*
.b) I
.130*
.456
.390*
.144+
.630
.496
Notes: LCIPC comparison is averageof the z-scoresfrom the letter comparison and pattern comparison measures.DSRT = digit-symbol reaction time
*p < .01.
PROCESS PURITY AND UNIQUENESS OF AGE EFFECTS
dissociation procedure in both a memory (Jacoby et al.,
1993)and an attention(Toth et al., 1995a)task were found
to be reliable, and therefore useful for individual difference
research involving correlations. Second, both controlled
processingestimateswere related to many other measures,
but the automatic processing estimates were largely unrelated to these measuresand to each other. Third, although
the qualitative distinction between controlled and automatic
processeswas supportedby the patternsof relationsamong
variables,we found little evidencefor a unique age-related
influenceon the estimatesof controlledprocessing.Finally,
there was a high degree of overlap of age-relatedvariance
in the controlled processingestimatesand the measuresof
simple processingspeed.We briefly addresseach of these
issuesin the following paragraphs.
Reliability: - One goal of the presentstudy was to examine the reliability of estimatesderived from the processdissociationprocedure.Assessingreliability is the first step
toward more comprehensivecomparisons of different
performancemeasures,and is also a prerequisitefor interpreting relations of the measuresto individual difference
variablessuch as age. Previousresearchwith the processdissociationprocedurehas shown that both aging (Jacoby
et al., 1997)and brain injury (Toth, in press)may affecr estimates of mnemonic control, but have no effect on estimates of the automatic processesinvolved in memory.
However,in the absenceof infbrmation about the reliability
of those estimates.it is conceivable that automatic estimates have weaker relations to variables such as age becausethey are less reliable than are estimatesof controlled
processing.The discovery in this study that both types of
process estimates from a memory paradigm and an attention paradigm had moderateto high reliability (i.e., from
.76 to .98) makes this artifact-basedinterpretationof the
different pattems of relations for controlled and automatic
processingestimatesunlikely.
Distinctiveness of process estimates.- A second major
finding of the study was that the estimatesof controlled
processing were related to many other measuresof performance, while the estimatesof automatic processinghad
weak to nonexistentrelations with these other measures.
This pattern is consistentwith the view that the controlled
measuresreflect deliberate strategic processesthat are involved in many cognitive tasks, whereas the automatic
measuresreflectprocessesthat are more task-specific(Neumann, 1984).The lack of a significantrelationshipbetween
the estimates of automatic processing in the memory and
attentiontasks(i.e., r = .09) is also compatiblewith the idea
that the nature of automaticity varies according to the particular task under investigation.
The closer correspondenceof the controlled than of the
automatic measuresto other cognitive variableswas also apparent in the results of the single common factor analysis.
That is, Figure 3 illustrates that the controlled measureshad
moderate to high loadings on the factor representingwhat
was common to the other measures,but there were either
weak (and negative)or nonexistentloadings of the measures
of automatic processingon the common factor. This pattem
P227
is important becausethe fact that the automatic measures
share little variance with other measuresis consistent with
the view that they reflect a qualitatively different type of
processingthan that involved in the other measures.
Uniquenessof age-related influences.- The perspective
that cognitive performance is determined by qualitatively
different processesis supported by the evidence just mentioned regarding the different pattems of relations to other
variables for the estimatesof controlled and automatic processing. However, there was no evidence in the current
study that the age-relatedinfluences on the controlled measures were unique, in the sensethat they were independent
of the age-relatedinfluenceson other cognitive measures.
Previous researchhas revealed that a broad range of cognitive variableshas negativerelations to age, but that few
of theserelationsare independentof one another(e.g.,Lindenberger& Baltes, 1994; Salthouse,1994).A similar outcome was evident in this study becausevariables representing verbal memory, verbal fluency, trail making, and
several measuresof speed were found to share large proportions of their age-relatedvariance.Moreover, as in previous studies.when the variableswere examinedto determine which of them had unique age-relatedinfluencesafter
taking into accountthe relation betweenage and what the
variableshave in common, the independentage-relatedinfluenceswere found to be few in numberand small in magnitude.The new linding in this study is that the estimatesof
controlledprocessingmay also sharethe sameset of causal
influencesas many other cognitive measuresbecausethere
is no independentage-relatedinfluenceon those measures
in the common factor analysis.
Patterns of mediation - As one would expect when
many measuresshare large proportions of age-relatedvariance,mediationalrelationscan be postulatedbetweenpairs
of measures.However, an interestingimplication from the
single common factor perspectiveis that measureswith the
strongestloadings on the common factor should have the
greatest impact as potential mediators of the age relations
on other measures.Examination of Figure 3 and Tables 3
and 4 revealsthat this was generallythe case,as the speed
measurestended to have the highest loadings on the common factor and were also the measureswhich, when statistically controlled, led to the greatestattenuation of the agerelatedvariancein other measures.An interestingquestion
for which we do not yet have an answer is whether this pattern is merely a statistical consequenceof the differential
loadings on the comrnon factor, or whether the measures
with the highest loadings are somehow more central or fundamental to the common factor.
Conclusion. - Many contemporary measuresof cognitive performance involve a mixture of theoretically distinct
processes.It is therefore important to disentanglethose processesand to identify their separatecontributions to relations between other variables such as age. Results from this
study demonstrate that reliable measures of theoretically
distinct processescan be obtained from two very different
cognitive tasks, and that these estimateshave different pat-
P228
SALTHOUSE
ETAL.
terns of relations to age and other cognitive measures.
However, in addition to trying to obtain process-puremeasures,it is also important to consider the extent to which
the age-relatedinfluences on different measuresare unique
and independent from the age-related influences on other
measures.That is, many measureshave been found to exhibit age-relateddifferences, but they may not all have
unique causes.Another finding in the presentstudy was that
measuresof controlled processingfrom both a memory task
and from an attention task sharedall of their age-relatedinfluenceswith a variety of other cognitive measures,suggesting that they may simply be additional manifestations
of a broader and more fundamental Dhenomenon.
AcxNowr-socvgNrs
This researchwas supportedby National Institute on Aging Grant R37
AG-06826 to Timothy A. Salthouse.We would like to thank Amy Cosby
and Renee Stein for assistancein the recruitins and testins of research
participants.
Address correspondenceto Dr. Timothy Salthouse,School of Psychology, Georgia Institute of Technology, 725 Atlantic Drive, Atlanta, GA
3O332-Ol7
O. E-mail: Tim.Salthouse
@psych.gatech.edu
RepenercEs
Benton,A. L., Hamsher,K. D., Varney,N. R., & Spreen,O. (1983).Judgment of Line Orientation, Form V. New York: Oxford University Press.
Cohen, J., & Cohen, P (1983). Applied multiple regression/correlation
analysisfor the behavioralsciences(2nd ed.). Hillsdale,NJ: Lawrence
ErlbaumAssociates.
Cowan, N. (1996).Can we resolvecontradictionsbetweenprocessdissociation models?Consciousness
and Coenition,5,255-259.
Curran, T., & Hintzman, D. L. (1995). Violations ofthe independcnceassumption in processdissociation.Journal of ExperimentalPsychology:
Learning, Memory, and Cognitktn, 2l , 531-547 .
Delis, D. C., Kramer, J. H., Kaplan, E., & Ober, B. A. (1987). California
Verbal lcarning la.sr.New York: Psychological Corporation.
Graf, P., & Komatsu, S. (1994). Processdissociationprocedure:Handle
with caution ! European J ournal of Cognitive Psltchoktgy, 6, I 13-129.
Hay, J. F., & Jacoby, L. L. (1996). Separating habit and recollecrion:
Memory slips, proiess dissociations, and probability matching. Journal of Experimental Psychoktgy: Learning, Memory, and Cognition,
22, t323-1335.
Jacoby,L. L. (1991). A processdissociationframework: Separatingautomatic from intentional uses of memory. Journal of Memory and ktnguage, 30, 5 I 3-541 .
Jacoby,L. L., Begg, I. M., & Toth, J. P. (1997). In defenseoffunctional independence:Violations of assumptionsunderlyingthe processdissociation procedure? Journal of Experimental Psychology: Learning,
Memory, and Cognition, 23, 484495.
Jacoby,L. L., Jennings,J. M., & Hay, J. F. (1996).Dissociaringautomatic
and consciously controlled processes:Implications for diagnosis and rehabilitation of memory deficits. In D. J. Herrmann, M. K. Johnson, C.
McEvoy, C. Hertzog, & P. Hertel (Eds.), Baslc and applied memory research:Theoryin context.Vol. l. (pp. 161-l 93). Hillsdale,NJ: Erlbaum.
Jacoby,L. L., Toth, J. P., & Yonelinas,A. P. (1993). Separatingconscious
and unconsciousinfluencesof memory: Measuringrecollection.,/ournal of Experimental Psychology: General, 122, 139-154.
Kliegl, R., & Mayr, U. (1992). Commentary on Salthouse(1992)'Shifring levels of analysisin the investigationof cognitive aging." Human
Development, 35, 343-349.
Kucera, H., & Francis,W. M. (1967). Computationalanalysis of present
day American English. Providence, Rl: Brown University Press.
Lindenberger,U., & Baltes, P B. (1994). Sensoryfunctioning and intelligence in old age. A strong connection. Psychology and Aging, 9,
339-355.
Lindsay, D. S., & Jacoby,L. L. (1994). Stroop processdissociations:The
relationship between facilitation and interference. Journal of Experimental Psychology: Human Perception and Performance, 20,219-234.
Neumann, O. (1984). Automatic processing:A review of recent findings
and a plea for an old theory. In W. Prinz & A. F. Sanders(Eds.), Cognition and motor processes(pp. 255-293). Berlin: Springer-Verlag.
Reingold, E. M., & Toth, J. P (1996). Processdissociationsversus task
dissociations:A controversyin progress.In G. Underwood (Ed.), Implicit cognition (pp. 159-202).Oxford: Oxford University Press.
Reitan, R. M. (1992). Trail Making Tesls.Tucson,AZ: Reitan Neuropsychology Laboratory.
Salthouse,T. A. ( I 993). Speedand knowledgeas determinantsof adult age
differences in verbal tasks. Journal of Gerontologlt: Pgychological
Sciences,48, P29-P36.
Salthouse,T. A. (1994). How many causes are there of aging-related
d e c r e m e n t si n c o g n i t i v e f u n c t i o n i n g ? D e v e l o p m e n t a lR e v i e w , 1 4 ,
4t3431 .
Salthouse,T. A. (1996). The processingspeedtheory of adult age diff'erencesin cognition.PsychoktgicaL
Review,103,403428.
Spieler,D. H., Balota, D. A., & Faust,M. E. (1996). Stroop performance
in healthy younger and older adults and in individuals with dementia
of the Alzheimer's type. Journal of Experimental Psychoktgv: Human
Percept ion and P erJbrmance, 22, 46 | 41 9.
Toth, J. P. (in press).Memory and attentionfollowing traumatic brain injury: Processdisscriations.The Clinical Neuropsychobgist.
Toth, J. P., Levine, B., Stuss,D. T., Oh, A., Winocur, G., & Meiran, N.
( I 995a). Dissociationof processesunderlying spatial S-R compatibility: Evidence fbr the independentinfluenceof what and where. Consciousnessand CoRnition, 4, 483-50 I .
Toth, J. P., Reingold,E. M., & Jacoby,L. L. (1994).Toward a redelinition
of implicit memory: Processdissociationsfollowing elaborativeprocessing and self-generation.Journal of Experimental Psyr:hobgy:
Leaming, Memory, and Cognition, 20,29(>303.
Toth, J. P, Reingold, E. M., & Jacoby,L. L. (1995b).A responseto Graf
and Komatsu'scritique of the processdissociationprocedure:When is
caution necessary?European Journal oJ Cognitive Psychoktgy, 7,
l l3-r30.
Woodard,J. L., Benedict,R. H. 8., Roberts,V J., Goldstein,F. C., Kinner,
K. M., Capruso,D. X., & Clark, A. N. (1997). Short form alternative
to the Judgment of Line Orientation Test. Journal of Clinical and Expe rimentaI N europsycholo gy, /8, 898-904.
ReceivedNovember 8, 1996
AcceptedApri1
l 7,1997