DevelopmentalPsychology Copyright

1993by theAmericanPsycholosicallAssmrldllijlb
1993.vol. 29. No. 4. 722-73E

in Cognition
SpeedMediationof Adult AgeDifferences
Timothy A. Salthouse

Previous researchhas establishedthat a large proportion of the adult age-relatedvariance in

various measuresoffluid or processcognition is reducedwhen statisticalcontrol proceduressuch
as hierarchical regressionare used to eliminate variation in measuresofperceptual speed.This
finding was confirmed in the presentstudy and was extendedto include paired-associateand
free-recallmeasuresof memory in addition to measuresof reasoningand spatialabilities. Most of
the speedmediation was associatedwith speedmeasuresrequiring cognitive operationssuch as
comparisonor substitution insteadof merelymotor responsessuchascopyingsymbolsor drawing
lines.Theseresultssuggestthat the rateof performing elementarycognitiveoperationsis an impor-
tant proximal mediator ofthe adult agedifferencesin severaltypes ofcognitive tasks.

Inspiredlargelyby ideasof Birren (e.g.,1955,1965,1974), ageR2 for agewhen it wasconsideredalonewas.158,but the

Salthouse0985a) proposed that the adult age differencesin valuewasreducedto only.035afterthe variationin a compos-
certain measuresof cognitive functioning might be partially ite measureof perceptualspeedwas controlled. Theseresults
mediated by age-relatedreductionsin the speedofexecuting thereforesuggestthat almost807o(i.e.,[.I 58 - .035]l I 58 = .778)
relevantmental operations.The evidenceavailableat that time ofthe age-related variancein certain measuresoffluid cogni-
wasratherweak,becauseno publishedstudieshad beenexplic- tion is associatedwith variationsin perceptualspeed.
itly designedto investigatethis hypothesis.Most of the data The researchreported in this article was designedwith two
relevantto the hypothesiswereobtainedby locating published major goalsin mind. One goalwasto examinethe generalityof
articlesreportingcorrelationsamongage,a measureof process- the speed-mediationphenomenon by including measuresof
ing speed,and a measureof cognition (or studiesby Salthouse memory in addition to measuresof fluid or processaspectsof
containing the relevant data), and then computing partial cognition.Extendingthe researchin this direction is ofinter-
correlationsbetweenageand the cognitivemeasureafter par- est, becauselessconvincingsupport for a mediationaleffectof
tialing out the measureof speed.Only a few data setswith the speedwasfound with m€asuresreflectingmemory functioning
necessaryinformation could be identified, and the samples in the partial correlation analyses reported in Salthouse
wereoften small and usually involvedtwo extremeagegroups (l 985a).That is, partialinga measureofprocessingspeedhad a
rather than a continuous range of ages.Furthermore, in all smaller effecton the correlationsbetweenageand measuresof
casesthe relevantconstructswere assessed with a singlemea- memory than it did on the correlationsbetweenageand mea-
sureratherthan with potentially more reliableand valid multi- suresof other types of cognitivefunctioning. Moreover,addi-
ple measures.Despite these limitations, the hypothesis re- tional analyseson the data from the Salthouseet al. (1988)
ceivedsomesupport becausestatisticalcontrol of the measure project similar to those summarized in Table I also revealed
ofspeed frequentlyresultedin an appreciablereduction in the weakerattenuationofthe agerelationsficrmeasuresrepresent-
magnitudeof the relation betweenageand the measureof cog- ing memory functioning than for measuresrepresentingother
nition. types of cognitive functioning. For example,the attenuation
In the last 5 years,severalmore extensivestudieshavebeen was only 63.8Vofor a measureof the accuracyof reproducing
conductedin which data relevantto the hypothesishavebeen the identitiesof items in a matrix, only 4'l .59ofor a measureof
reported.The major onesinvolving at least 200 adultswith an the accuracyof reproducingthe positionsof items in a matrix,
agerangeofat least35 yearsand including severalmeasuresof and only l6.7vofor a measureof paired-associate memory.
both speedand cognition are summarized in Table 1. In all
The smaller attenuation of the age-related memory differ-
cases,the availablespeed measureshave been combined to
encesafter control of measuresof speedraisesthe possibility
form a more reliablecompositeindex. (This requireda reanaly-
that quicknessof mental operationsmay be a less important
sisofthe data from Salthouse,Kausler,& Saults,I 988) It can be
factor in the agedifferencesin traditional memory teststhan in
seenthat resultsfrom thesemore recentstudiesareclearlycon-
cognitivetestsassessing reasoningor spatialabilities. Although
sistentwith the speed-mediationhypothesis.That is, the aver-
this speculationis plausibleon the basisofthe availableresults,
strongerevidenceofthe differential influenceofspeed on cog-
nitive and memory measuresis desirablebeforeone can con-
This researchwas supportedby National Institute on Aging Grant
clude that speedfactors have a minimal role in the relations
R37 AC06826.I thank JenniferShawfor valuableassistance in super-
vising the testingof participants and the initial analysesof the data. betweenageand memory.One type of evidencethat would be
Correspondence concerningthis article shouldbe addressedto Tim- relevant in this regard would be a discovery that substantial
othy A. Salthouse,SchoolofPsychology,GeorgiaInstitute ofTechnol- age-relatedeffectsremained after statistical control of a mea-
ogy,Atlanta,Georgia30332-0170. sure of perceptual speed, or equivalently,a finding that the
722
 SPEED MEDIATION 723
relations between age and measuresof memory were atten- verted into the seconds-per-itemmetric and wasanalyzedin a
uatedonly slightlyafter oneremovedthe variancein the percep- linear regressionequation relating those times to the average
tual speedmeasuresthat waslinearly relatedto the measuresof times of the referencegroup on the sametasks.It is important
memory.Severalmemory measureswerethereforeincluded in to emphasizethat unlike most previousreports of systematic
the presentstudy to allow this implication to be investigated. relations(but seeCharness& Campbell,1988,for an excep-
The secondmajor goalof the presentstudy wasto examinea tion), these analysesare conducted at the level of individual
variety of different speedmeasuresto identify the aspectsof subjects.That is, a regressionequationwascomputed for each
speedmost responsiblefor mediating age-cognitionrelations. subjectrelating his or her times (asthe dependentor criterion
Two issueswereof particularinterestin this connection.One variables)to the mean times of the studentreferencegroup (as
wasthe extentto which sensoryor motor factorswereinvolved the independentor predictor variables).
in the influenceofperceptualspeedon age-cognitionrelations. Linear regressionequationscan be characterizedin terms of
Becausemost testsofperceptual speedinvolvevisual presenta- three parameterscorrespondingto the correlation, the inter-
tion of materialsand requirea written responseto communi- cept, and the slope.The correlation is informative about the
catedecisions,sensoryand motor aspectsmay be contributing degreeto which there is a systematic(linear)relationamongthe
to many measuresof perceptualspeed.In an attemptto remove variables.The intercept indicates the magnitude of an influ-
the influenceofperipheral input and output speedwhen analyz- encethat is constant acrossall variables,and the slope repre-
ing the influenceofperceptualspeed,participantsin the pres- sentsthe amount by which the individual'stime increasesfor
ent researchproject, in addition to performing more conven- everyunit increasein the time of the reference group.Of these
tional perceptualspeedtests,also performed severalspeeded parameters, the slopeis often consideredthe most interesting,
taskswith minimal cognitiverequirements. becauseit hasbeeninterpretedasa reflectionofthe magnitude
Anotherissuerelatedto the measurement of speedconcerns of age-related slowing(e.g.,Cerella,I 985,199I ; Cerella,poon.&
the significanceof the systematicrelationsoften reported be- Williams, 1980;Lima et al.,199I ; Myersonet al., I 990).
tweenthe averagetimes ofyoung and old adultsacrossdifferent A strong interpretationof the systematicrelationsamong
combinationsof speededtasks.Many researchershave found speedmeasuresis that the slopeis a direct index ofa general
that thereis frequentlyan orderlyrelationbetweenthe response speedfactor. The slope is not measuredin absoluteunits be-
timesof youngadultsacrossa rangeofexperimentalconditions causeit is alwaysrelativeto somereferencegroup,but it can still
or tasksand the responsetimes of older adultsin thosesame be meaningfulin across-individual comparisonsif everyoneis
conditionsor tasks(e.g.,Brinley,1965;Cerella,1985,l99l; evaluatedagainstthe samereference group.A weakerposition
Lima, Hale,& Myerson,199l; Myerson,Hale,Wagstaff,poon, regardingthe systematicrelationsamong speedmeasuresis
& Smith,1990;Salthouse, 1985a,I 985b).Ir is possibleto disrin- that the slopeis merelyan indexof the relationamongvariables
guishbetweenstrongandweakpositionsregardingthetheoreti- that areeachdeterminedby otherinfluencesin additionto the
cal significance of thesesystematicrelations.The strongposi- hypothesizedspeedfactor. Becausethe contribution of the
tion emphasizes the quantitativeform of the relationbetween other influencescan be expectedto differ acrossvariables.the
the times of youngand old adultsbecauseof the assumption weak positionviewsthe slopeas a lessdirect index of the hy-
that the relationis informativeabout both the magnitudeand pothesizedspeedfactorthan manyofthe observedspeedmea-
the natureof the age-related slowingevidentacrossmany mea- sures.(SeeSalthouse,1992a.for further discussionof this dis-
suresof speededperformance.A weakerposition maintains tinction)
that the existenceofsystematicrelationssuggests that the age- In summary,in both the strongand weak interpretations, a
relatedinfluenceson the variousmeasuresare not indepen- high correlationbetweenthe times of the individual and the
dent, but this perspectivedoes not necessarilyconsiderthe timesof the reference groupimpliesthat the measures areasso-
lorm of the relationinformativeabout either the rate or the ciatedand not independent,but the meaningofthe slopepa-
precisenatureofage-related slowing.That is, lack ofindepen- rameteris quite diffbrentin the two perspectives. The strong
dencecan be inferredifknowledgeofthe agedifferences in one position maintainsthat the slopeis an important index of a
measureprovidesinformation about the age differencesin centralspeedconstructbecauseit has a specialstatusas the
other measures, but no specialsignificanceis attachedto the mostdirect indicatorofthe hypothesized speedfactol whereas
parametersof the functionsrelatingthe times of youngadults the weak positionviewsthe slopeas much lessinterestingbe-
to the timesof olderadults. causeit is assumedonly indirectly to reflectthe operationof
The distinction betweenstrongand weak positionsregard- any relativelygeneralage-relatedspeedfactorsthat might exist.
ing the systematicrelationsbetweenthe timesof adultsof dif- Onemannerin whichthe validityof thesealternative concep-
lerentagescan be elaboratedafter one first considershow these tualizationsof the systematicrelationsamongspeedmeasures
relationsweredeterminedin the presentstudy.The initial step might be investigated involvesexaminingthe strengthof link-
involvedthe administrationof a batteryof speedtests-involv- agesbetweenvariousmeasures ofspeedand both ageand cog-
ing a rangeof operationsand completiontimes-to 100college nition.The reasoningis asfollows.If, asthe resultssummarized
studentswho servedas the referencegroup.Scoreson eachof in TableI suggest,speedof processing is an importantmediat-
thesetestswerethen convertedinto unitsof secondsper item, ing factorin the agedifferences in somemeasures of cognition,
with the averages rangingfrom 0.4 s to almost2 s per item.The then comparisonsof the degreeto which age-cognitionrela-
nextstepin the procedureconsistedof administeringthe same tions are attenuatedafter control ofalternative speedmeasures
batteryof speedteststo a sampleof over300 adultsfrom a wide may be informativeaboutthe statusof thosespeedmeasures as
rangeofages.Eachoftheseparticipants'scores werethen con- reflectionsof a common or generalspeedfactor.Specifically,
724 TIMOTHY A. SALTHOUSE

TableI
Resultsof Studies Examining the Relations Betwem Age and Cognition Beforeand After
Statistical Control of PerceptualComparison Speed
Age R2 for age
range
(in years) Speed r(age - speed) Cognition Before After 70attenuation

Schaie(1989)
6ll 22-84 FindA's, IDPict -.53 PMA Space .1 6 8 .036 78.6
PMA Reason .281 .058 79.4
628 22-84 FindA's, IDPict -.65 PMA Space .22r .032 85.5
PMA Reason .360 .058 83.9

Hertzog ( 1989)
592 43-78 FindA's, IDPict ? PMA Space .124 .01l 9r . l
Number Compar. PMA Reason .237 .02'l 88.6
AnswerSpeed Spatial Relations .209 .032 84.7
't7.3
SpatialVisualization .128 .029
Induction .202 .02r 89.6
(l99la)
Salthouse
221 20-80 Letter Compar. -.61 Shipley# Cor. .221 .024 89.l
PatternCompar. Raven # Cor. .322 .056 82.6
Shipley7oCor. .l l'7 .o2l 82.r
Raven 7oCor. .t29 .ol2 90.7
Working Memory .292 .050 82.9
228 20-82 Letter Compar. -.7| PapFld # Cor. .037 .007 8l.l
PatternCompar. CubAsm # Cor. .019 .002 89.5
Analog # Cor. .1 8 9 .001 99.5
Reason# Cor. . 165 .022 86.7
PapFld 7oCor. .069 .006 91 . 3
CubAsm 7oCor. .056 .003 94.6
Analog 7oCor. .t23 .ol2 90.2
Reason7oCor. .t2l .015 87.6
Working Memory .254 .014 94.5
223 20-84 ktter Compar. -.60 PapFld # Cor. .IJJ .029 78.2
PatternCompar. CubAsm # Cor. .097 .033 66.0
Analog # Cor. .227 .0t2 94.'l
Reason# Cor. .1 8 6 .020 89.2
PapFld 7oCor. .203 .065 68.0
CubAsm 7oCor. .1 7 8 .070 60.7
Analog 7oCor. .083 .009 89.2
Reason7oCor. .104 .03'l 64.4
Working Memory .208 .012 94.2

Salthouseand Babcock ( I 99 I )
233 l 8-82 Letter Compar. -.65 Working Memory .2tl .006 9't.2
PatternCompar.

Salthouse,Kausler,and Saults( 1988)

ZJJ 20-78 Digit Symbol .46 Analog Time .328 .100 69.5
Number Compar, SeriesCompl.Time .100 .019 81.0
Analog 7oCor. .1 8 4 .092 50.0
SeriesCompl.7oCor. .o'14 .033 55.4

and Mirchell(1990)
Salthouse
383 r 8-84 FindA's -.28 PapFld # Cor. .l'12 .t37 20.3
Number Compar. SurDev # Cor. .179 .128 28.5
lrtSet # Cor. .lt2 .046 58.9
Shipley# Cor. . r08 .049 54.6
PapFld 7oCor. .l16 .082 29.1
SurDev 7oCor. .04'7 .025 46.8
LetSet7oCor. .014 .001 92.9
Shipley7oCor. .018 .008 55.6

Note. FindAs=FindingAsTestfromEkstrom,French,Harman,andDermen(1976);IDPict=IdenticalPicturesTestfromEkstrometal.(1976);
PMA Space= SpaceTest from Primary Mental Abilities Battery (Schaie,1985);PMA Reason= ReasoningTest from Primary Mental Abilities
 SPEED MEDIATION 725
Tablel bontinued\

Battery(Schaie,1985);NumberCompar.: NumbercomparisonTestfrom Ekstromet al. (1976);shipley#

cor. = numberof correctresponses in
ShipleyAbstractionTest;Shipley%Cor..= percentageofresponsescorrect in ShipleyAbstractionTestjRaven#Cor.: numberofcorrectresponses
R3y"n: AdvancedProgressiveMatrices Test; Riven 7oCbr. = percentageof rlsponsescorrect in Raven's progressive
iL Advanced Matrices Test;
Working Memory = compositeof ListeningSpanand Computation Spanicores;PapFld # Cor. : number
of correct responsesin a paper lolding
test; PapFld% Cor' = percentageofcorrect responsesin a paperlolding test;CubAsm# Cor. : number
ofcorrect responsisin a.uU. urr'",oUtv,.rt;
CubAsm.ToCor. : percentageof correct responsesin a cube assembly-test; Analog # Cor. = number of correct responsesin a geometricanalogres
test; Analog 7oCor. - percentageofcorrect responsesin a geometricanalogiestest; Reason# Cor. = number
ofcoirect respon-ses
in an integrative
reasoningtest;Reason% Cor.= percentage ofcorrectresponses in an integfativereasoningtestgAnalojiime : mediantime percorrectresponse
in a geometricanalogies :
test;seriescompl.Time mediin tim. p.. "or..!t ,.rponsein a si.ier "o-pt.iion t.it; ieriescompl.vo
Cor.-percentage
of correctresponses in a seriescompletiontest;SurDev# Cor. = numberof correctresponses in tire SurfacebevelopmentTest(Ekstromet al.,
I 976);SurDev7ocor. = percentage ofcorrect responsesin the SurlaceDevelopmentTesi(Eksrromet al..I 976);Letset# cor. - numberofcorrect
responses in the LetterSetsTest(Ekstromet al.,I 976);LetSet7oCor.= percentage ofcorrectresponses in the LetterSetsTest(Ekstromet al.,| 976).

the attenuationproducedby statisticalcontrolofthe slopepa- Method

rametershouldbe greaterthan that producedby statisticalcon_
trol ofthe observedmeasures ifthe slopeis a directreflectionof Subjects
the speedfactor,but it should be lessthan that producedby at
Demographiccharacteristics of the individualswho participatedin
leastsomeof the observedmeasures if the slopeonly indirectly this project are summarizedin Table 2. Collegestudentsreceived
representsthe influenceof a common speedfactor. The pro_ credit toward a course requirementfor their participation, and
posedmethodfocuseson relationswith cognitionto discrimi_ membersof the adult sample,who were recruitedlrom newspaper
nateamongthe alternativeinterpretations ofspeedbecauseof advertisements, received$10 lor their participation.Becauseof the
the assumptionthat speedis interestingprimarily becauseof relativelyhigh meanlevelsofeducation, theadultsamplecanbeconsid_
the implicationsit hasfor other aspectsof cognitivefunction- eredto be a positivelybiasedsamplecomposedlargelyof peoplefrom
ing. middleand high socioeconomic levels.Healthstatuswasevaluatedby
The presentresearchprojectinvolveda batteryoflT tests, responsesto three questions:(a) How would you rate your overall
yielding l9 separatemeasures,that was administeredto two healthon ascalefrom l: excellent165= poorwith3 = average?, (b)
How many prescriptionmedicationsdo you take eachweek?;and (c)
samplesof researchparticipants.One sampleconsistedof 100
How many times in the past 5 yearshaveyou been hospitalizedor
collegestudentswho servedas the referencegroup for certain receivedother treatmentfor cardiovascular or neurologicalproblems
analyses, and the othersampleconsistedof 305 adultsranging (e.g.,heart attack,stroke,or high blood pressure)? As can be seenin
from l9 to 84 yearsofage.Elevenofthe testswereintendedto Table 2, the number of medicationsand medical trearmentswas
measurespeed,with 8 of them speciallydesignedto vary sys_ greaterwith increasedage,but the overall health ratingsremained
tematicallywith respect1othe amountof requiredprocessing. relativelyconstant.(SeeSalthouse, Kausler,& Saults,I 990,for similar
These8 new taskswere composedof four pairs, with each lindings)
memberof the pair requiringsimilar activitiesbut involving
slightlydifferentmaterials.The four pairsinvolvedthe follow-
ing operations:Marking (i.e.,crossingvertical or horizontal Procedure
linesto makeplussigns);Copying(i.e.,copyinglettersor num- All testingwas conductedin small groups,and everyparticipant
bers);Completing(i.e.,comparingtwo stringsof lettersor digits receivedthe testsin the samefixedorder.Materialsfor the testswere
and then writing the item missingfrom onestring);and Trans_ assembled in folders,whichalsocontainedthe informed-consent form
forming(i.e.,incrementingan initial valueby one or two in the and a brief questionnairerequestinginformationabout age,gender,
alphabeticor numeric sequenceand then writing the new health,and yearsofeducationcompleted.The orderofthe testswasas
value).Threeotherspeedmeasures werealsoincludedto facili- follows:Digit Symbol,Line Marking(in theordervertical.horizontal.
tate comparisonswith earlierresearchusing thosemeasures: vertical,and horizontal),Copying(in the ordernumbers,letters,num-
bers,and letters).Completion(in the ordernumbers,letters,numbers,
the WechsIer I d uIt I nteIIigence ScaIe- Revrserl(WAIS-R) Digit
and letters),Transformation(in the order numbers,letters,numbers,
SymbolSubstitutionTest(Wechsler, 198I ) and the LetterCom- and letters).PatternComparison(in the order three-itempairs, six-
parisonand PatternComparisonmeasures describedby Salt_ item pairs, and nine-item pairs), Letter Comparison(in the order
house(l 99I a) and Salthouse and Babcock0 99 I ). three-itempairs,six-itempairs,and nine-itempairs),pMA Reasoning
The criterion,or to-be-explained, variableswerelour cogni- (Schaie,1985),PMA Space(Schaie,1985),integrativereasonrng, geo-
tive measures and four memorymeasures. The cognitivemea- metricanalogies, paired-associate
memory,and free-recallmemory.In
sureswerederivedfrom testssimilar to thoseusedin earlier all but the memorytests,the participantswereencouraged to perlorm
studies.Thesewerethe Primary MentalAbilities(pMA) Space asrapidlyand asaccuratelyas possible.Instructionsin the two mem-
and PMA Reasoningtestsusedby Hertzog(1989)and Schaie ory testsemphasizedtrying to rememberas many wordsas possible.
(1989)and the integrativereasoningand geometricanalogies with guessingencouraged when in doubt abouta response.
testsusedby Salthouse(l99la). The four memory measures The Digit Symbol,PMA Reasoning, and pMA Spacetestsweread-
ministeredaccordingto the publishedinstructionsexceptthat answers
werederivedfrom a paired-associates test and a free-recalltest wereto bewrittendirectlyon the testlorm in the two pMA testsrather
yielding measuresof recall accuracyfor the primacy,asymp-
than on separate answersheetsasspecifiedin the standardadministra-
tote,and recencysegments of the serialpositionfunction. tionsofthesetests.This modificationwasintroducedto avoidpossible
726 TIMOTHY A. SALTHOUSE

Table 2
Means and Standard Deviations of DemographicChamcteristics
of the ResearchSamplesby Age Decade
Medical
Age Education Health Medication treatment
Age
decade M ^sD M .tD M SD M ,SD M SD

Men
Students 59 20.9 1.3 13.3 1.3 t.7 0.8 0.27 0.78 0.08 0.53
20s 20 23.9 3.0 l4.t t.9 2.3 1.4 0.20 0.52 0.00 0.00
30s l9 34.4 2.9 1 5l. 2.9 2.0 r.3 0.r l 0.32 0.00 0.00
40s t t 43.9 2.9 1 5l. 2.5 2.7 1.6 0.53 1.23 0 .l 8 0.53
50s 2l 54.2 3.5 15.8 3.4 2.5 t.2 0.62 l.16 0.10 0.30
60s JJ 65.0 2.2 16.4 l.9 2.6 1.3 1.52 t.82 0.52 1.06
70s l4 74.2 3.7 16.1 2.9 2.4 t.4 l .00 t.4l 0.93 2.64

Women
Students 4l 20.8 t;l t3.7 1.5 2.0 r.0 0.29 0.5| 0.00 0.00
20s 22 24.2 2.6 15.0 t.4 2.3 1.3 0.55 0.86 0.18 0.39
30s 22 34.5 3.0 t5.7 t.7 2.5 t.4 0.32 0.78 0.86 3.23
40s 36 44.9 2.8 15.4 2.4 2.3 1.3 0.50 l.l3 0.17 0.56
50s J/ 53.9 3.0 15.8 2.6 2.1 1.3 1.22 t.2t 0.09 0.39
60s 39 64.6 2.7 14.7 2.7 2.3 1.2 1.44 2.t2 0 .1 3 0.41
70s 30 74.0 3.7 14.0 2.3 2.4 |.2 1.63 1.27 0.97 2.'t7

Note. Education refersto yearsof formal educationcompleted,and health representsa self-assessment

on a 5-point scaleranging from | = excellentto 5 = poor. Medication refersto number ofprescription
medicationstaken eachweek,and medical treatmentrefersto numberof treatmentsfor cardiovascularor
neurologicalproblemsin the past 5 years.

confusionor delaysassociatedwith the useofa separateform for re- box, nothing in the bottom box, and either one or two +'s (pluses)in
cording one'sanswers,which may be greaterwith increasedage(e.g., the middle box. The plus signs in the middle box indicated how the
Bellucci & Hoyer,1975;Hertzog, 1989;Hoyer,Hoyer,Treat, & Baltes, letter or number in the top box wasto be transformed;one plus meant
l 978-l 979). that the value should be incrementedby one in the alphabeticor nu-
Two versionsof the Line Marking test wereadministered,one with meric sequence,and two plusesmeant that the valueshould be incre-
28 rowsofl6 horizontallinesand the otherwith 23 rowsof20 vertical mentedby two. For example,an F with two pluseswould yield H, and a
lines. Researchparticipantswereinstructed to work from left to right 5 with I plus would yield 6. After carrying out the transformation
placing short vertical lines through eachhorizontal line and to work specifiedin the middle box, the participant must write the identity of
from top to bottom placingshort horizontal lines through eachverti- the new letter or number in the bottom box. Each page contained
cal line. The pageswerearrangedsuchthat vertical marks weremade sevenrows ofl5 setsofboxes, and the separatelytimed (30 s) pages
first, then horizonlal marks,then vertical marks, and finally horizon- were presentedin the order numbers,lett€rs,numbers,and letters.
tal marks.The time allowed to work on eachpagewas30 s. The Pattern Comparison and Letter Comparison tests were very
The Copying test wasadministeredin two versions,one containing similar to the earlier versionsofthese testsdescribedby Salthouseand
Babcock (1991).The patterns were constructed by connecting dots
numbersand the other letters.In both cases,the test pagescontained
within an imaginary 4 X 4 matrix, and the letter sequenceswerecon-
l0 rowsofl 5 pairs ofboxes, with a letter or number in the top box and
structed by random selection (without replacement)of consonants.
nothing in the bottom box. The participant'stask wasto copy the ilem
Within eachtest, there werethreeseparatelyiimed (30s)sections,with
in the top box in the empty box immediatelybelow it. There werefour
successive sectionscontaining pairs ofitems with three, six, and nine
separatelytimed sectionsin the task, involving numbers,letters,num-
elementsper item. The task for the participant wasto examinethe two
bers.and letters.with 30 s allowed for eachsection. "S" on the line betweenthem if
membersof the pair and to write an
The Completiontestswerealsoadministeredwith both numberand they were the sameand a "D" on the line if they were different. One
letter versions.The materials in these tests consistedof pagesof 50 half of the pairs on eachpagewere the same,and the other half were
pairs of numbers or letters, with one element missing from one
different becauseofan alteration in the identity or the position ofone
memberof the pair. The completememberof the pair containednine of the elements.Each sectionof the test consistedof 32 pairs of line
elementsand appearedrandomly on either the left or the right. The patternsor 64 pairs of letter strings.'
participant'stask wasto examinethe completememberof the pair to
determine the identity of the missing elementand then to write that
elementin a box correspondingto the number or letter missing from I Analyseswere also conductedon the Letter Comparisonand Pat-
the other member of the pair. There were four separatelytimed sec- tern Comparison scoresaccording to the number of items (letters or
tions in the task, involving numbers,letters,numbers,and letters,with line segments)beingcompared.Correlationsbetweenageand the aver-
30 s allowed for eachsection. age of the Letter Comparison and Pattern Comparison scoreswere
The Transformationtest wassimilar to the Copying test exceptthat -.58 for three items,-.54 for six items,and -.52 for nine items.The
eachitem consistedofthree boxes,with a letter or a number in the top percentageattenuationoftheage-cognition relations,analogousto the
 SPEED MEDIATION 727
The integrativereasoningtest consistedof72 problems,eachcon_ ity ofthe averagevalue.Becausethere wasonly oneadministration for
taining premisesdescribingthe relationbetweentwo terms(e.g.,R and sometestsin this study,reliability could not be computedfor all mea_
S, do the OPPOSITE),and a questionabout the consequences on one sures.However,reliabilities ofstandardized testshavebeen reported
term of a specifiedchangein another term (e.g.,If R DECREASES, in other sources,with valuesof .82 lor the Digit Symboltest (Wechsler,
will S INCREASE?).(SeeSalthouse, Mitchell,Skovronek,& Babcock, I 98I ), .72to.8 I for the PMA Spacetest(Schaie, I 9g5),and.g2 to.g6 firr
I 989;and Salthouse,Legg,palmon,& Mitchell, 1990,lor further de_ the PMA Reasoningtest (Schaie,1985).
scription of this task) One problem within eachsetof three problems Most ofthe valuesin Table3 appearsatislactorybecausethe reliabil_
had a singlepremisebelore the question,one problem had two prem- ity estimatesare generally in the moderate to high range of .g0 or
ises,and one problem had three premisespresentedbeforethe ques_ greater.The three measuresfrom the free-recalltaskare notableexcep_
tion. The correct answ€rfor half of the problemswasyes and that for tions,perhapsbecausethesecomponentsoffree-recall memoryperfor_
the remaining problemswas ro. In all cases,the task for the research mance are not very stable when assessedwith only two trials. Esti_
participantwasto placea mark eitherin a column labeledyesor in a
mated reliability of a measureof totar recail acrossalr serial positions
column labeledno. Participantswereallowed4 min to completeas was somewhatgreater (i.e., .68), but the component measureswere
many problemsas possible. retainedbecausetheyareofgreatertheoreticalinterestthan the undif_
Thegeometricanalogies testwasarrangedin a fashionsimilarto the ferentiatedmeasureof total recall.
integrativereasoningtest in that therewere72 problems,with I ofeach The completecorrelationmatrix with all measures, includingage,is
of 3 problemtypes in everyset of 3 problems.problemsin this test presentedin Table4. All of the correlationsinvolvingagein the main
consistedof four boxes,with eachbox containingbetweenone and sampleof adults (abovethe diagonal)are negative,and most of the
three lettersin variousconfigurations.For example,one problem con_ remainingcorrelationsin this sampleare moderatein magnitude.
sistedof the followingarrangement of letters:first box_a solidletterC Correlationsin the sampleof students(belowthe diagonal)aresome_
in the upperleft corneranda solidletterA rotatedclockwise9ff in the what smallerand, as expected,becauseof the restrictedage range,
bottom right corner;secondbox-an outline letterC in the upperleft noneofthose involvingageweresignificantlydifferentfrom zero.
cornerand a solid letter A rotatedclockwise9f in the bottom right
corner;third box-a solid letterD in the upperleft cornerand a solid
letterC rotatedclockwise9f in the bottom right corner;lourth box_ MeasurementModel
a solid letter D in the upper left corner and a solid letterC rotated
clockwise9f in the bottom corner.The participant'staskwasto deter_ Relationsamongthe variableslor the main sampleof adultsin Table
mine if the patternof relationships betweenthe lettersin the first two 4 wereexaminedby meansof a confirmatoryfactoranalysisspecifying
boxeswasidenticalto that in the secondtwo boxes.The answerto the four factorsof motorspeed,perceptualspeed,memory,andcognition.
problemjust describedwould thereforebe no, becausethe shift of The hypothesizedfactor structure, with the estimatedcoefficients,is
solid to outline form for the letter in the upper left corner was nor portrayedin FigureI . Notice that a correlationwaspostulatedbetween
represented in the secondpair ofboxes.An equalnumberofproblems the residualsofeach ofthe measures derivedfrom parallelmethods,
had one,two,or threelettersin eachbox in the problem,and one half with the exceptionofthe two copy measures. Becausethe initial esti-
of the problemshad correspondingrelationships, and the other half mate of the correlation betweenthe residualsof the letter copy and
did not. As in the integrativereasoningtest,decisionswerecommuni_ number copy measureswas negative,no relation was specified be-
catedby placinga mark eitherin a columnlabeledyES or in a column tweenthe residualsof thosemeasures. The chi-squarefor the model
labeledNO. Participantswerealloweda time limit of 4 min to com_ illustrated in FigureI was121142, N: 305)= 3 19.25, p < .0t, indicating
pleteas manyproblemsas possible. that there wasa significantdiscrepancybetweenthe model and the
The words used in the two memory testswere four-letternouns data. However,other measuressuggestedthat the model provided at
selectedfrom the most frequent1,000wordslisted in the Thorndike leastan adequatefit to the data. Accordingto Steiger(19g9,p. 9l),
and Lorge(1944)word count. The rateofpresentationin both tasks valuesabove.90of the adjustedpopulationgammaindex,which is a
wasapproximatelyI word every 2 s. Two separatetrials with different coefficient of model determination adjusted for model complexity,
wordswerepresented in eachtask.Trialsin the paired_associates task and valuesbelow .10 of the adjustedroot-mean-square, which is a
consistedof the oral presentationof 6 word pairs.followedimmedi_ root-mean-square standardizedresidualadjustedfor model complex_
atelyby instructionsto turn to a pagecontainingthe first membersof ity,can beconsidered asrepresenting a goodfit. Valueswith this model
eachpair. Participantswereinstructedto write the secondmemberof were .928 firr the adjustedpopulation gamma index and .062 lor the
eachpair adjacentto the first term, and they wereencouraged to guess adjustedroot-mean-square. The Joreskog-Sorbom goodness-of-fitin_
evenifthey werenot sureofthe correctresponse. Approximately60 s dexwas.905.
wasallowedfor the response phaseofthe trial. Trialsin the free_recall A surprisingfeatureofthe coefficients in FigureI is the highcorrela_
taskinvolvedtheoral presentation of a list of l2 wordslollowedimme_ tion (.88I ) betweenthe latentvariablesofperceptualspeedand cogni_
diatelyby the wordRECALL, at whichtime the participantturnedthe tion. As a meansof testingwhetherthis correlationwassignificantly
pageand wrote as many wordsas he or shecould remember.Approxi- differentfrom 1.0,the path betweenthesetwo variableswas fixedat
mately60 s wasallowedfor the responsephaseofeach trial. 1.0and the fit of the model redetermined.The chi-squarefor the ad_
justed model wasx2(143,N: 305) : 362.22,p < .01, which yields
a
significantdifferencechi-squareof 42.9j (d.f: I, p < .01), indicating
MeasurementReliability that the altered model fit the data significantlyworsethan the original
Estimatesof the reliabilitiesfor the dependentmeasures model.Because the modelconstrainingthe correlationto 1.0provides
in the two
samplesare reported in Table 3. All estimateswere derived by using a significantly poorer {it to the data than the model with the correla_
the Spearman-Brownlormula to adjust the correlation betweenthe tion asa freelyestimatedparameter,it can be inlerred that, given the
valuesfrom the two administrationsofeach test to predict the reliabil_ assumptions represented by the structureofthe model,the true corre-
lation betweenthe perceptualspeedand cognition constructsin these
data is not equalto | .0. In other words,although the two consrructsare
highly related,they are not completelyidentical.
valuesreported in Table 8, were94.6Vofor three items. gg.67olor six The correlationbetweenthe motorspeedand perceptualspeedcon-
items,and 85-l7ofor nine items. structswas.8 I 2. The nestedmodel procedurewasusedto test whether
728 TIMOTHY A. SALTHOUSE

Table 3
Reliabilities of the PerformanceMeasures
Estimatedreliability

Measure Students

Horizontal mark: No. of horizontal marks made on vertical lines (30 s) .82 .88
Vertical mark: No. of vertical marks made on horizontal lines (30 s) .89 .92
Copying,letters:No. ofletters correctlycopied (30 s) .88 .97
Copying, numbers:No. of digits conectly copied(30 s) .88 .9)
Letter completion:No. of correct completionsof letter strings(30 s) .88 .86
Number completion:No. of correctcompletionsof number strings(30 s) .'7| .92
Letter transformation:No. of correcttransformationsof letters(30 s) .89 .92
Number transformation:No. of correct transformationsof numbers(30 s) .92 .91
ktter comparison:No. of correct same/differentjudgments about letter
strings(30 s) .83 .83
Pattern comparison:No. of correct same/differentjudgments about line
patterns(30 s) .85 .90
Digit symbol: No. of items correct in WAIS-R Digit Symbol Substitution
Test (90 s) (.82r
PMA Reasoning:No. of items correct in PMA SeriesCompletion
ReasoningTest (6 min) (.84)o
PMA Space:No. of items correct in PMA Figure Rotation Test (5 min) (.'7't)'
Integrativereasoning:No. of items correct in integrativereasoningtest (4
min)
Analogy: No. of items correct in geometricanalogiestest (4 min)
Pairedasociates:No. ofwords correct in paired associates
Primacy: No. of words recalledfrom serialpositions I to 4
30
.39
n
.5v
Asymptote:No. of words recalledfrom serialpositions5 to 8 .44 .50
Recency:No. of words recalledfrom serialpositions9 to 12 .)t .46

Note. WAIS-R = WechslerAdult IntelligenceScale-Revised; PMA = Primary Mental Abilities.

' Valuefrom Wechsler(198I b Averagevaluesfrom Schaie(l 985).
).

this correlationwassignificantlydifferent from 1.0by constrainingthe evidentin eachmeasure:The averageperformanceof adultsin

correlationto 1.0and examiningthe chi-squaredifferencebetweenthe their 60s and 70s is betweenone and two standarddeviations
original and alteredmodels.The differencechi-squarewas199.75(df= belowthat ofadults in their 20sand 30sfor eachofthe compos-
I , p < .01), implying that the correlationwassignificantlylessthan I .0. ite variables.
All standard errors for the coefficientsillustrated in Figure I were The secondnoteworthyfeatureofthe data in Figure2 is that
lessthan .06, and those for the coefficientsbetweenperceptualspeed
even the performanceof the adults in their 20s is appreciably
and cognition and betweenmotor speedand perceptualspeedwere
lower than that ofthe referencegroup ofstudents, particularly
both .024.Becausethesevaluesindicatethat the relationsbetweenthe
observedvariablesand the latent constructsare significantly greater for the cognition compositevariable.This differencemay be
than zero and that the correlationsbetweenconstructsare all signifi- attributableto the fact that the studentswere recruited from a
cantly less than 1.0, the constructs can be consideredto have both relatively select university and consequentlymay have had
convergentand discriminant validity, respectively(cf. Anderson & higher levelsofcognitive ability than the averageyoung adult
Gerbing,1988). recruitedfrom newspaperadvertisements to participatein proj-
ectsof this type. Somesupport for this interpretation is avail-
able in the observation that the students in this study per-
Results formed between 0.6 and 0.9 standard deviations above the
meansof the 22- to28-year-oldsin the normativesamplefor the
Composite Variables
1985venion of the PMA Spaceand Reasoningtests(Schaie,
Compositevariableswere createdfior eachof the forrr con- 1985).The fact that the nonstudentresearchparticipantsin
structsby averagingthe unit-weightedz scoresfor the measures their 20shad a slightly lower averagenumberofyears ofeduca-
hypothesizedto representeachconstruct(accordingto the rela- tion than adults in subsequentdecades(seeTable 2) could also
tions portrayedin FigureI ). Correlationsamongthesecompos- be interpretedasindicating that the 20- to 30-year-oldadultsin
itesand estimatesof their reliabilitiesare presentedin Table5. this sample were not as positively biased as the adults from
Notice that the correlations among the composite variables other ageranges.
havethe samepattern but are generallysmaller in magnitude Whether it is becausethe samplein the decadeof the 20s is
than thoseamong the latent constructsportrayedin Figure 1. lessselectthan the samplesfrom other agegroupsor because
Mean compositescoresat eachdecade,expressedin standard there are little age-relatedeffectsin thesemeasuresuntil mid-
deviationsof the scoresof the referencestudent sample,are dle age,the data in Figure 2 suggestthat performancein some
illustrated in Figure 2. There are two particularly salient fea- testsmay remain relativelystableuntil the 30sor 40s.The na-
turesofthese data.The first is the very pronouncedagetrends ture of the agerelationsin the compositevariableswasinvesti-
 SPEED MEDIAIION
729

al
? e q q : t q { { { { { E t + { i l iI \o€
orO
o\ a--
^; -i

T : f f E E i q { E s q q E { 3 r\IE i
a t , i
r 6 I O
olo

€ t <
? { q E q E E S d q q q q \ { li il Q ! nq
I
t-- o\
arO 6i O
l o
Q c - . 1 €q
r i . l f i € si fVO - ir q
r o r\ q q i q q I E
|€ q A 99 O\ \O a,
N _ r; -".;

E { \ n E q q i q q q 6 h qt qq q e \ qn
d d
td' :

E { E € n € q E n q q nt4{ E
\?q6 \* \no hcl
z
oi €; at

E r f f q { 6 q q q q q qt q
E;?tiq - f
r -
(d

dci
N :

i q q q q q E E q q qt tqq q : q : { o9 cl
€ r d + q
N

N
T { q q q q i 6 q q tqq i { 9 e a c E nrl
\oo 6i ci
r <

? q q q q E q \ q t{ q q s n q 6 t € e \r on aq .'i
N I

n E i q n q q q nt q q E f i n q l q e i q?
ol a{
6 r
r;6i
o

i l q q q € H EIqE q q i E " r q - € - e O\n

A.; I 9 o0

o
{ q E n q q qt h q { q ! ! E n : : ; q .o r;
\r nn
o
q \ , r q E E q q , r q q i . ' r q q{ fi i :
I N O
q \ rag
€ N
o

q!!qq l
I
tqiEqqtfreEtIi€: nq
N 6
\oo\l
6^i I
I I

E q E St E q q q q q q ? E q q € i l qn-n \o\ol
I

NO\ cj r l
\ o l
I

? E Er q q q a q f r q q l : : q E t ++ €
I
s € l o
d $
J.d I
h l
Su I
T Ar q E 4 { E q t \ q q i q i l : q € t r; -'
€ 6 1
cidl
e \o- r l
I
T r \ q E q i e E E q q s l € q E : €r ^ ii 6 i r o l
-io I
e - l
a

I I
F

I
^aJ
r ? ? ? € q € E ? E E q " ? : ?'-:E
€ ?qr
9 q n lI
N I
9)

- , :=
oo
\
\ P
q F . QE =
c.-= H E tr o
'=
x
\ o
tP , v b" cE E F . - b E sE F , t
ShEEX '= fr H
o
trEeE:oF*58F-" o : o
s E E E E E U E s E 5 E EHg !g S =
A t l : ! g : I g : ; & - d . r i i a
g
. . E F * - E E E E E E . ; < < F p € E EsEi a
T -I
O ! !
S r s g g*5l E E B E E ! € E f ;i g ^ ; s
= i \
-jdid + ri6 r dd €: - E: -
= = d: = 3 9 i 5i- a
- - H - N
730 TIMOTHY A. SALTHOUSE

.7Gr (

I
.586
I
(

3 4 (

.313 (

MEMORY

COGNITION

Figure l. Hypothesizedmeasurementmodel and coefficientsobtained from confirmatory factor analy-

sis.(MSPD : motor speed;PSPD= perceptualspeed.Variablesin the boxesare asfollows:HMark : no.
of horizontalmarks madeon vertical lines;VMark: no. of vertical marks madeon horizontallines;
LCopy = no. of letterscorrectly copied; NCopy = no. ofdigits correctly copied; LComp = no. ofcorrect
letter-stringscompletion; NComp = no. of correct number-stringscompletion; LTran = no. of correct
transformationof letters;NTran = no. of correct transformationof numbers;PatCom = no. of correct
same/differentpattern comparison judgments; LetCom = no. of correct same/differentletter compari-
sons;Digsym = digit symbol;PairAsoc= paired associates; PMAReas= Primary Mental Abilities Induc-
tive ReasoningTest; IntReas= integrativereasoning)

gatedmoresystematicallyby examiningthe significanceofqua- fect other than age was the main effect of education on the
dratic (age-squared)terms in the regressionequation.The qua- memorycompositevariable,,F0, 301): 10.79,p < .01,M,S =
dratic term was associatedwith a significant increment in R2 0.367, indicating that people with more educationp€rformed \
for the motor speedand memory variables,but in both casesit better on the memory tests than people with lesseducation.
was relativelysmall (i.e.,.029 and.037, respectively).Because
the variance associatedwith the linear age trend was much
The absenceofany interactionsinvolving agesuggeststhat the I
age trends in the compositevariableswere not moderatedby
larger(i.e.,R2valuesrangingfrom . 184to .297),only the linear gender,availableindexesof health, or education,and conse-
relationswereconsideredin all subsequentanalyses. quently thesefactorswere ignored in all remaining analyses.
Main effectsand interactionsof gender,self-assessed health
status,numberof medications,numberof medical treatments, Hierarchical Regression Analyses
and yearsofeducation werealsoexaminedin regressionequa- Resultsof the hierarchicalmultiple regressionanalysescon-
tions on the four compositevariables.The only significantef- ducted on the compositevariablesare summarizedin Table6.
 SPEEDMEDIATION
731
Table 5
sociatedvariancein the cognition measurefrom.202when age
CorrelationsAmong Composite Scores
wasthe only variablein the equationto.003 when the variance
Variable associatedwith perceptualspeedwascontrolledby enteringage
in the regressionequationafter perceptualspeed.
t . Age -.46 -.55 -.43 -.45 Second,the earlier resultsare extendedby the finding that
.I 2. Motor speed
3 . Perceptualspeed
.04
.08
(.e5)
.59
.70
(.97)
.38
.51
.43
.74
the relationsbetweenageand perceptualspeedarestill statisti_
4. Memory .05 .16 cally significantafter controlling the influenceof motor speed
.23 (.77) .52
5 . Cognition -.01 )\ st (.86) (i.e.,R2 of.062) and by the finding that therewasa largerattenu_
,l Note. Values
,J I
ation ofthe age-relatedeffectsin cognition after control ofper_
abovethediagonalofnumbersin parentheses arecorre_ ceptualspeedthan after control of motor speed(i.e.,reductions
lationsfromtheadultsample(r: 305),andthoie belowthediagonal
arecorrelations lromthestudentsample(z= I 00).Values r1 R2 to .003 for perceptualspeedvs. .0g0 for'motor speed).
in pare-nthe_
sesarereliabilitiesof the composite scoresfor the adultsampleesti- This combination of resultssuggeststhat only a portion of the
matedbytheformuladescribed by Kenney (1979,p.132):.ef iaUifity: age-relatedslowingthat afects cognitivefunctioning is motoric
n(average r)/lt + (n- t)(averagerjl. in nature.
Third, anothernew finding is that the age-relatedvariancein
measuresof memory is also substantiallyattenuatedafter sta_
Entriesin the first column of this table indicate the cumulative tistical control of perceptualspeed.The percentageattenua_
R2 in the prediction of the criterion variableafter the variable tion,82.6Vo,is not as largeasthe 98.52oevidentin the cognitive
on that row and the variableson immediately precedingrows measure,but it is still quite substantial.
had been enteredinto the regressionequation.Entries in the And fourth, agewasassociatedwith a very small proportion
secondcolumn indicatethe incrementin R2associatedwith the of the total variance in the cognitivemeasureafter the influ_
addition ofthe variableon that row into the regressionequa_ encesof motor
speed, perceptual speed, and memory were
tion. Finally, valuesin the third column indicate the F values taken into consideration.Furthermore, the direction of the
fo-rthe significanceeitherof the initial R2or of the incrementin standardizedregressioncoefficientfor agein the completere_
R2. gressionanalysiswasactually positiverather than negative,in_
Severalpointsshouldbe noted regardingthe resultssumma_ dicating that if anything,
increasedage was associatedwith
rized in Table6. First, it is clearthat earlier findings wererepli_ higher levelsof cognitiveperformancewhen the contributions
catedconcerningthe attenuationofthe relationsbetweenage of motor speed, perceptual speed, and memory were con_
and cognitionafter statisticalcontrol ofa compositemeasureof trolled.
perceptualspeed.This is evidentin the reductionofthe age_as_
The resultsjust describedare relevantto both of the maior

0.5 Memory
U' --c"-
E
c MSpd
f 0 -a-
c PSpd
o -0.5 \ l ---l---
(U \ Cognition
'5
o -1
o
o
(u - 1 . 5
o
c(u
U) -2
c
o
\ o -2.5
f
I F
U)
-3
20 30 40 50 60 70
Chronological
Age
Figure 2' Mean performanceat each decadefor the four composite
variablesexpressedin standard
deviationsofthe t00 collegestudentsserving as the referencegroup. (Bars
aboveor below the symbols
representone standarderror. MSpd = motor speed;pSpd = perceptual
speed)
732 TIMOTHY A. SALTHOUSE

Table6 ceptual speedand cognition, respectively.Coefficientsin the

Hierarchical RegressionResultsBasedon top panel of Figure 3 are basedon the standardizedregression
CompositeScoresin Aduh Sample coefficientsfrom multiple regressionanalyseson the composite
scores.Coefficientsin the bottom panel werederivedfrom the
Criterion/predictor R2 Incr. R2
EZPNIII (Steiger,1989)structural equationprogram with the
Motor speed latentconstructsdefinedin the measurementmodel illustrated
Age .2t4 82.45* in Figure l. (That is, this figure illustratesthe structural coeffi-
Perceptualspeed cientsfrom a structural equation model superimposedon the
Age .297 t28.t2* measurementmodel representedin Figure l) The chi-square
for this modelwasy21l6S,N: 305): 414.92,p < .01,but the
Motor speed .495 337.30*
Age
goodness-of-fitmeasuresindicated an adequate fit of the
.557 41.80*
model to the data (i.e.,adjusted population gamma index :
Memory .907,adjustedroot-mean-square : .069,and Joreskog-Sorbom
Age .r84 68.I 2* goodness-of-fit index = .883).Explorationofseveralvariations
Motor speed .t47 57.58* ofthe model portrayedin Figure3 revealedthat the fit could be
Age .22'7 .080 3 l . 3 l* improved by minor alterationsof the structure representedin
Perceptualspeed .261 tt2.24r FiguresI and 3 (e.g.,specifyingdirect as well as indirect paths
Age .293 .032 13.70* betweenageand the letter transformation,numbertransforma-
Motor speed .t47 62.69+ tion, and recencyvariables), but noneofthe modificationsre-
Perceptualspeed .262 .ll5 48.7l* sultedin an appreciablechangein the structuralparameters
Age .293 .031 13.20* depictedin the bottom ofFigure 3.
Cognition
Age .202 76.68i
Motor speed .1 8 5 75.98*
(A)
Age .265 .080 32."18*
Perceptualspeed .546 365.20+
Age .549 .003 2.08
Memory .275 I 25.35+
Age .337 .o62 2 8 .I 5 +
Motor speed .1 8 5 t28.5"1*
Perceptualspeed .562 .377 26l .80+
Age .567 .00s 3 . 7|
Motor speed .185 I 36.78+
Perceptualspeed .562 .3-17 2 7 8 . 5I *
Memory .593 .031 23.1t*
Age .594 .001 1.05

Note. Incr.R2: incrementin R2association

with additionofsecond.
third. or fourth variable.
* p <.01.

goals ofthe project. They addressthe issueofspeed mediation

of age differencesin memory, becausestatistical control of the
composite measure ofperceptual speed greatly attenuatesthe
age-relatedvariance in the composite measureof memory. The
resultsare also relevantto the issueofthe type ofspeed involved
in the mediation, because the attenuation of the age-related
variance was greater after control of measures of perceptual
speed than after control of measuresof motor speed.

Structural Model
A structural model illustratingthe hypothesizedrelations
amongthe primary constructsis represented in Figure3. The
causallinkagesare hypothesizedto be from motor speedto
perceptualspeedand from memory to cognition rather than Figure 3. Hypothesized structural relations and path coefficients
vice versa,becausemotor speedand memory areassumedto be basedon the compositevariables(A) and on the latent variables(B).
more elementarythan, and perhapsevenconstituentsof, per- (MSPD: motor speed;PSPD: perceptualspeed)
 SPEED MEDIAIION 733
It is apparentin Figure3 is that althoughthe relationbetween Table 7
perceptualspeedand cognition is highly positive, there is a Resultsof Commonality AnalysesConductedon the Cognition
negativerelationbetweenmotor speedand cognition.This was and Memory CompositeVariqblesin Adult Sample
unexpected,becauseit indicatesthat when perceptualspeedis
controlled, fastermotor speedis associatedwith lower levelsof Predictor variable
cognitive performance.Interpretation of this finding should
Motor Perceptual
probably be deferred until it has been confirmed in another Criterion Age speed speed
sample,but it may reflectthe operationof sometype of impul-
siveness.In most other respects,however,the relations por- Cognition
trayed in Figure 3 are similar to the conclusionsinferred from Unique to age .005
Unique to motor speed .018
the regressionanalyses.Specifically,thereare negativerelations Unique to perceptualspeed .302
betweenageand motor speed,perceptualspeed,and memory; Common to ageand motor
no significant relation between age and cognition; and large speed -.002 -.002
positiverelationsbetweenmotor speedand perceptualspeed Common to ageand
perceptualspeed .0'74 .074
and betweenperceptualspeedand cognition. Moreover,it is
Common to motor speed
reassuringto note that the samepatternsare evidentwhen the and perceptualspeed .045 .045
analysesare basedon unit-weightedcompositescores(Figure Common to age,motor
3A')and on latent constructsdefined by a confirmatory factor speed,and perceptual
analysis(Figure3B). speed .124 .124 .t24
Total effects .202 .1 8 5 .545
The results from the structural analysesare related to the Memory
primary goals of the project in the following manner. First, Unique to age .031
examination of the paths leading to the memory variable re- Unique to motor speed .000
vealsthat some,but not all, of the age-relatedvariancein mem- Unique to perceptualspeed .065
Common to ageand motor
ory is mediatedthrough slowerperceptualspeed.That is, un-
speed .001 .001
like the cognitivevariable,a direct path existsbetweenageand Common to ageand
memory in addition to the indirect path through perceptual perceptualspeed .049 .449
speed.And, second,at leastwhen motor speedis assumedto Common to motor speed
influenceperceptualspeed,the relationsbetweenmotor speed and perceptualspeed .044 .044
common to age,motor
and cognition are very small and are actually negativerather speed,and perceptual
than positive,whereasthe relationsbetweenperceptualspeed speed .103 .r 0 3 .r 0 3
and cognition are largeand positive. Total effects .1 8 4 .148 .261

Commonality Analysis
Another method of partitioning the variance in a criterion ever,this analysisextendsthe earlieranalysesby indicatingthat
variableis commonalityanalysis(Pedhazur,1982).As applied the largestproportion of variancein the cognitionand memory
in the presentcontext, the goal of this techniqueis to decom- compositevariableswas sharedamong age,motor speed,and
posethe total effectsof ageon either the cognition or memory perceptualspeed.
compositevariableinto a unique contribution of ageand into
contributionsin common with motor speed,perceptualspeed,
Relations Among Speed Measures
or both. The unique influencecorrespondsto the age-related
variancethat is independentofboth motor speedand percep- As describedin the introduction, regressionequationswere
tual speed,and the common influence representsthe age-re- computed relating the times of each adult participant to the
latedvariancethat is sharedwith either motor speedor percep- averagetimes of the 100studentsacrossthe samemeasuresof
tual speedor both. speed.That is, the predictor or X valuesin the regressionequa-
Table 7 contains the summary information from the com- tionswerethe meantimesof the 100students,and the criterion
monality analysesconducted on the cognition and memory or )z valueswere the times of the individual subject in those
compositevariables.Entries in the first column (Age)are of same tasks.Intercept, slope,and correlation coefficientsindi-
greatestinterestbecausethey indicatethe partitioningofthe cating the goodnessof fit of the linear equationweretherefore
age-relatedeffects.Two important points should be empha- obtained for eachsubjectto representthe relation ofhis or her
sizedabout thesedata. First, notice that only a small propor- times to the mean times of the referencegroup.
tion ofthe total age-relatedeffectsis uniquely associatedwith Meansof the regressionparameterswere .93 (,SD: .07) for
age(i.e.,.005vs..202for the cognitionvariableand.03l vs..184 the correlation,I .40 (SD : 0.9I ) for the slope,and -0. I 3 (.SD:
for the memory variable).And second,noticethat the variance 0.54) for the intercept.The high averagecorrelation, together
common to ageand perceptualspeedis much larger than that with the fact that the magnitudeof the correlation coefficient
common to ageand motor speed(i.e.,.074 vs. -.002 for the wasnot significantlyrelatedto age(i.e.,r - -.14), indicatesthat
cognition variableand .049 vs. .001 for the memory variable). the linear equationsprovided a reasonablecharacterizationof
In both respects,these resultsreinforce the conclusionsfrom the dataof most participants.The interceptparameterwassig-
the hierarchicalregressionand path analysisprocedures.How- nificantly relatedto age(i.e.,r = - .19), but the absolutemagni-
734 TIMOTHY A. SALIHOUSE

tudeofthe effectwasquitesmall,becauseeachyearofage was predictor in the regressionequationbut that the increment in

associatedwith a reduction in the intercept of only .006. The R2 associatedwith ageafter controlling the influenceof motor
slopeparameterhad a signihcantpositiveagerelation(i.e.,r: speedwas.080.Dividing the differenceof .122(i.e...202 - .0S0)
.30)correspondingto an annualincreaseof.0l7. by .202yields.604,which whenmultipliedby 100resultsin the
Comparableanalyseswerealsoconductedon only the 7 per- first entry in Table8.
ceptualspeedmeasures, excludingthe 4 motorspeedmeasures. The recencymemory measureis excludedfrom Table 8 be-
Resultsof these analyseswere similar to those basedon all causecomparisonsof the magnitudesof attenuationwith this
speed measuresexcept that the correlation coefficient was measurewould not be meaningfulgiventhe very small relation
intercept: -. I 3; slope: 1.45),suggesting
smaller(i.e.,r: .821. a betweenageand this measure(i.e.,r = -.08). Entrieslabeted
somewhatpoorer fit of the linear equation to the data. The Averagein the last row and in the last column are included to
correlation betweenthe slopebasedon all I I speedmeasures facilitatecomparisonsacrossmeasures,but it shouldbe empha-
and that basedonly on the 7 perceptualspeedmeasureswas sizedthat theseare not true averages, becausethe valuesin the
.98. columnsand rowsare not independent(e.g.,the compositecog-
To estimatethe reliabilityof theslopeparameterbasedon all nition variableis derivedfrom the variableslabeledpMA Rea-
I I speedmeasures,the speedmeasureswere rank-orderedin soning,PMA Space,integrativereasoning,and analogy).
terms of averagetime and then separateslopescomputed for It can be seenin Table8 that statisticalcontrol ofthe compos-
the six odd-rankedmeasures (i.e.,horizontalmarking,number ite perceptualspeedmeasuregenerallyresultedin the greatest
copying,numbertransformation,patterncomparison,number attenuationof the age-relatedeffects,followed by a composite
completion,and letter comparison)and for the five even- basedon the LetterComparisonand PatternComparisonmea-
ranked measures(i.e.,vertical markings,letter copying,digit sures,and then by the Digit Symbolmeasure.The composite
symbol, letter completion, and letter transformation).The motor speedmeasurewas relativelyineffectivein attenuating
meansfor the two slopeswere1.42and 1.38,respectively, and the agerelationson anyof the cognitiveor memorymeasures,
the correlationwas .48. Applicationof the Spearman-Brown and the two slopemeasuresfared the worst in terms of their
formulato this correlationresultedin an estimatedreliability inferred mediationalinfluenceon the age-cognitionor age-
of the slopebasedon all I I speedmeasures of .65. memory relations.
Table8 containsa comparisonof the magnitudeofthe attenu- Theseresultsare relevantto the secondmajor goal of the
ation ofthe agerelationsacrossdifferentspeedmeasures and project in that they are informative about the relative impor-
acrossdifferentcognitiveand memorymeasures. Valuesin the tanceof differentmeasuresof speedas mediatorsof the rela-
tableweredeterminedby computingthe R2associated with age tions betweenageand cognitionor memory.Basedon the data
whenit wasthe only variablein the equationand againwhenit containedin Table8, one can inler that perceptualspeedis
wasenteredafterthe variableslistedin the top row ofthe table. much more important than motor speedand that although
The differencebetweenthe two varianceestimateswas then there is a systematicrelationbetweenthe speedsof adultsof
divided by the total age-related variance,and this value was dilferentages,the slopeof this relationdoesnot appearto be
multipliedby 100.To illustrate,Table6 revealsthat the R2for very importantasa mediatorof age-cognitionor age-memory
ageon the cognitionvariablewas.202when agewasthe only relations.

Table8
Attenuatkn oJAge-CognitionRelalionsas a F-unctionof the Measuresol
Speedand Cognitionin the Adult Sample

70 attenuation ofR2 for age

Motor Perceptual Digit

Criterion speed speed Lettcr/pattern symbol Slope pSlope Average

Cognition 60.4 98.5 95.5 89.I 46.0 3t.7 10.2

Memory 56.5 82.6 80.4 73.4 34.2 22.8 58.3
PMA Reasoning 68.3 100 98.8 94.6 54.5 38.3 75.8
PMA Space 43.9 "10.3
89.2 87.8 39.2 2'/.7 59.7
Integrativereasoning 55.6 '19.3
93.3 88.9 34.1 2t.5 62.1
Analogy 75.4 94.2 98.6 100 59.4 39.1 77.8
Pairedassociates 51.2 80.2 15.9 /J.) J l.) 2l.0 55.6
Primacy 57.5 73.2 68.6 59.5 32.7 24.2 52.6
Asymptote 42.5 66.4 60.4 57.5 18.7 I 1.9 42.9
Average 56.8 86.4 83.9 17.5 38.9 26.5 6t.7
Nole. Motor speedis a compositeof horizontalandverticalline markingand letterand numbercopying;
perceptualspeedisa compositeof letterand numbercompletion,letterand numbertransformation, letter
and pattern comparison,and digit symbol;letter/patternis a compositeof letter and pattern comparison;
slopeis the linear regression
sloperelatingtimesof I I speedmeasures to the averagetimesof t 00 college
students;PSlopeis the linearregressionsloperelatingtimesof7 perceptualspeedmeasures to the average
timesof 100collegestudents.PMA : Primary MentalAbiliries.
 SPEEDMEDIATION 735
General Discussion measurecould be interpretedas suggestingthat the factorsre-
sponsiblefor the recencyeffect(e.g.,primary memory) may be
T heoretical Construct s
relativelyunafected by increasedage(e.g.,Craik, 1977).How-
Although the four constructsof primary interestin this proj- ever,resultsfrom anothersimilar project are inconsistentwith
ect-motor speed,perceptualspeed,memory,and cognition- this interpretation.Ronnberg(1990)alsoadministeredl2-word
were ficund to have moderate to large correlationswith one lists to adults from a wide rangeof agesand analyzedperfor-
another,they arenevertheless distinguishable.That is, the mea- mancein terms of recall of wordsfrom successive thirds of the
surementmodel illustrated in FigureI wasfound to providean list. Correlationswith agein that study were-.12, -.35, and
acceptablefit to the data, and the correlations between the -.27 for the primacy,asymptote,and recencysegments,respec-
compositesin Table 5 were appreciablylower than the esti- tively,comparedwith the valuesof -.39, -.37. and -.08 in this
matedreliabilitiesof the composites. study The Ronnbergstudy thereforefound smaller age-related
The discovery of high correlations(i.e.,.74 for composites influenceson recall of primacy items compared with recency
and.88 for factor scores)between the perceptualspeedand items,whereasthe reversewastrue in this study.Becausepartic-
cognitionvariableswassomewhatsurprising,becausetheseare ipants in the Ronnbergstudy receivedeight lists of words and
generally assumed to represent separateand distinct con- participants in the presentstudy receivedonly two lists, the
structs, but correlationsof similar magnitude have been re- discrepancyin the two setsof resultsmay be relatedeither to
ported in other studies.For example,in studiesinvolvingsam- the reliability of the measures(which would be greater with
ples ofolder adults,the correlationsreported betweenpercep- more observations)or to sometype of age-differentialstrategy
tual speedand inductivereasoningcompositeswere.78(Baltes, shift acrosssuccessive lists.
Cornelius,Spiro,Nesselroade, & Willis, 1980),.80 (Cornelius,
Willis, Nesselroade, & Baltes,1983),and .78 (Schaie,Willis, Speed Mediation
Hertzog,& Schulenberg,l9ST). Schaie,Willis, Jay,and Chipuer
(1989)havealso reporteda correlationof.86 betweenpercep- Consistentwith the resultsof the studiessummarizedin Ta-
tual speedand inductivereasoning(and.77 betweenperceptual ble I , the age-relatedvariancein the measuresof cognitivefunc-
speedand spatialvisualization)in a sampleof 1,621adultsbe- tioning wasfound to be greatlyreducedafter statisticalcontrol
tween22 and 95 yearsof age.A lateranalysisof thesedata with of an index of perceptualspeed.In fact, the resultswith the
a total of 1,628adults resultedin correlationswith perceptual compositevariablesindicatethat between807oand 1007oof the
speedof .88 for inductivereasoningand.78 with a spatialfactor age-relatedinfluenceson memory and cognition were elimi-
(Schaie,Dutta, & Willis, 199I ). The resultsof all thesestudies nated by using statisticalproceduresto equateparticipantson
thereforesuggestthat there is considerableoverlapin what is an index ofperceptual speed.Expressedsomewhatdifferently,
measuredby perceptualspeedtestsand timed testsof inductive 18.470 of the total variancein the memorycompositeand20.22o
reasoning and spatial visualization, particularly in samples of the total variancein the cognitioncompositewereassociated
containinga largeproportionofolder adults. with agebeforestatisticalcontrol ofthe perceptualspeedvari-
The motor speedand perceptualspeedconstructsalso ap- able,but after statisticalcontrol ofperceptualspeed,the age-as-
pear to be distinct despitefairly high correlationsbetweenthe sociatedvariancewasonly 3.27ofor the memory compositeand
factors(.812)and the compositevariables(.70).Schaieet al. 0.3Vofor the cognitioncomposite.
(199I ) havereportedsimilar resultswith a more complex mea- Althoughthe resultsof this studydo not by themselves indi-
sureofmotor speedbasedon the speedofcopying paragraphs catehow speedmediatesthe influenceofage-relatedeffectson
printed in uppercaseand lowercaselettersand on the speedof memory and cognition, it is neverthelesspossibleto speculate
writing synonymsand antonymsof familiar words.The corre- about the mechanismsthat might be involved.An initial as-
lation betweenthis motor speedfactor and a perceptualspeed sumption is that perceptualspeedmeasuressuch as those in-
factor in their study was.9l3, and, as in the presentstudy, cluded in this study reflectthe rate at which the individual can
analysesrevealedthat this valuewassignificantlylessthan 1.0. carry out many elementarymental operations.Given this as-
sumption,one can speculatethat at leastsomeof the agediffer-
Age Relations ences in simple cognitive measuresmay be a direct conse-
quenceof a slowerspeedof mental processingbecausethe re-
The resultssummarized in Figure 2 and in Tables4 and 5 quired operations are relatively undemanding. Much of the
clearlyindicatethat increasedageis associatedwith slowerper- age-relatedinfluenceson the cognitivemeasuresin the present
formanceon manyspeededtasksand with lowerlevelsof perfor- study may be attributable to this type of influence, because
manceon certain memory and cognitivetasks.The agecorrela- most of the items in thesetests would probably be answered
tions with the compositemeasures(seeTable 5) ranged from correctly by individuals of moderateto high ability working
-.43 to -.55, indicatingthat between187oand 307oof the total without time constraints.In other words,with testsof low diffi-
variance in thesemeasureswas systematicallyrelated to age. culty,a largeproportion ofthe variancein performancecan be
These valuesare in the same range as those summarized in hypothesizedto be causedby variationsin the numberof items
Table I and are consistentwith much researchin the area of attempted rather than to variations in the percentageof at-
agingand cognition (seeSalthouse,199 I b, for a recentreview). tempted items answeredcorrectly.
The correlationbetweenageand the memory compositewas With more complex cognitivemeasures,the speedinfluence
-.43, but the age correlation was only *.08 for the recency may be largely indirect and perhapsmediated by an impair-
measure.The lack of a significantagerelationwith the recency ment in the functioningof working memory.The precisenature
736 TIMOTHY A. SALTHOUSE

of the workingmemoryimpairmentis still not clear,but it may terns with thesemeasuresin this studyweregenerallysimilar to
be relatedto a diminishedability to maintain the productsof thoseof the earlierstudies.
earlyprocessing duringthe executionoflater processing. Much Becauseit was recently reported (Salthouse,1992b)that a
higherorder thinking involvingintegrationand abstractioncan very large proportion of the age-relatedvariance in the Digit
be postulatedto requirethe simultaneousavailabilityof rele- Symbolmeasurewassharedwith a perceptualspeedcomposite
vant information,and the amountof simultaneously available createdfrom the Letter Comparisonand PatternComparison
informationis likely to be a direct functionof the effectiveness measures,the present data were examined to determine
of workingmemory.Threeobservations areconsistentwith this whetherthis phenomenonwas alsoevidentin this study The
working-memorymediation interpretation.First, the influ- earlier results were replicated becausethe R2 for age in the
enceof workingmemoryhasbeenfoundto be greaterthan that prediction of Digit Symbol performancewas.261 when it was
ofperceptualspeedfor difficultcognitivetestssuchasthe Ship- consideredalonebut wasonly .019whenconsideredaftersta-
ley AbstractionTestand the RavenbAdvancedProgressive Ma- tisticalcontrolof the LetterComparison/Pattern Comparison
trices Test (StudyI in Salthouse,1991a),whereasthe reverse composite.Although the residualagerelationwasstill signifi-
wastrue for easiercognitivetestssuchasthe integrativereason- cantly greaterthan zero, it was only 7Vaof the original value,
ing and geometricanalogiestestsincludedin this study(Stud- indicating that approximately93Voof the age-relatedvariance
ies 2 and 3 in Salthouse,l99la). Second,a similar patternof in the Digit Symbolmeasurewassharedwith a compositeof the
greaterworking-memoryinvolvementand reducedinfluence two perceptualcomparisonspeedmeasures.
of perceptualspeedin moredifficultor complexconditionshas
beenfound in within-taskcomparisons. That is, moredetailed
analyses ofthe resultsfrom Studies2 and 3 in Salthouse Comparison of Speed Measures
(l 99 I a)
revealedthat the influenceof working memory was greater The discoverythat there was more attenuationof the rela-
than that of perceptualspeedfor the more complexitems in tions betweenageand cognition after statisticalcontrol of per-
eachof the tests(Salthouse, 1992c).And third, the resultsof ceptualspeedmeasuresinvolvingcomparisonor transforma-
severalindependentstudiesindicatethat the age-related vari- tion operationsthanaftermotorspeedmeasures requiringlittle
ancein measures of working memory is greatlyattenuatedby or no cognitiveoperationssuggests that the speedof primary
usingstatisticalcontrolproceduresto removethe varianceas- interestin the mediationof relationsbetweenageandcognition
sociatedwith perceptualspeed(Salthouse, l99la. 1992d;Salt- is perceptualor cognitivein natureratherthan motoric.Also
house& Babcock,l99l). This suggests that speedmay be an consistentwith this interpretationis the finding that, at least
importantmediatorof the agedifferences in at leastsomemea- for the sampleof adults involvinga wide rangeof ages,the
suresof working memory. relationsbetweenthe compositemeasureswere,if anything,
It is not yet obviouswhat factorsbesidesperceptualspeed slightlystrongerbetweenperceptualspeedand cognition(r:
contributeto the agedifferencesin memory functioning.One .74) thanbetweenperceptualspeedand motor speed(r: .70).lt
possibilityis that working memory playsan important role, thereforeappearsthat the aspectofprocessingspeedresponsi-
much like that hypothesized with complexor difficult reason- ble for the mediationof agedifferences in cognitionis not that
ing or spatialproblems.Anotherpossibilityis that someof the associated with sensoryor motor processes but ratheris related
agedifferences areattributableto variationsin the effectiveness to the speedwith which simplecognitiveor mentaloperations
of particular mnemonicstrategiesor in the elficiencyof spe- can be executed.
cihc processes concernedwith encoding,retrieval,and so on. Statisticalcontrolof the slopeparameterled to only a small
Systematicinvestigationof the effectsof perceptualspeedon attenuationof the age-cognitionrelations,particularlyin com-
adult agedifferencesin specilicmemory componentsand on parisonwith other speedmeasures. As discussedin the intro-
theefficiencyofselectedstrategies is obviouslyneededto distin- duction,an outcomeof this type is more consistentwith the
guishbetweenthesepossibilities. weakinterpretationof theslopeparameterin whichit is consid-
eredto be an indirect reflectionofthe speedfaclor involvedin
age-cognitionrelations.The existenceof the systematicrela-
Comparabi litv Wilh Previous Studies tions among the speededvariablesimplies that the various
speedmeasures arenot independentwith respectto age-related
Becausesomeof the sameperceptualspeedand cognitive influences,but the presentdataofferlittle supportfor the view
taskshavebeenusedbefore.the presentresultscan be com- that the slopeis a direct indicatorofthe speedfactorpostulated
paredwith thoselrom the previousstudies.The attenuationof to be involvedin mediatingsomeof the agedifferences in cog-
the agedifferences in integrativereasoningwith statisticalcon- nition. The lact that the estimatedreliabilityof the slope(i.e.,
trol of the LetterComparison/Pattern Comparisoncomposite .65)waslowerthan that for the otherspeedmeasures reported
in this study was 88.97o, and the correspondingvaluesin the in Tables3 and 5 raisesthe possibilitythat the weakerattenua-
two relevantstudiesreportedin Salthouse(1991a)were86.77o tion of the agerelationswith the slopemeasuremight be attrib-
and 89.2VoValuesfor the analogiestask were 98.6%in this utableto the lowerreliabilityof this measurerelativeto that of
study,and 99.5Vo and94.lVoin the two earlierstudies.Hertzog the directly observedmeasures.To investigatethe extent to
(l 989)and Schaie(l 989)useddifferentmeasures of perceptual which differentialmeasurementreliability contributedto the
speedin their studies,and thusthe valueswith the PMA Space resultssummarizedin Table8, correctionsfor attenuationwere
and PMA Reasoningtestscannot be compareddirectly,but appliedto the correlationsbetweenageand speedand between
inspectionofthe valuesin TablesI and 8 revealsthat the Dat- speedand the compositemeasureof cognition.Thesedisatten-
 SPEEDMEDIATION 737
uatedcorrelationswerethen usedin the computationofpartial chophysiologyand speedof response.American Psychologist,29,
correlations between age and cognition after controlling the 808-8l 5.
relevantmeasureof speed.Finally, to expressthe resultsin a Brinley,J. n 0965). Cognitivesets,speedand accuracyofperformance
form comparablewith that of Table8, the squareof the partial in the efderly.In A. T, Welfiord& J. E. Birren (Eds), Behavior,aging
correlationwassubtractedfrom the squareofthe age-cognition and the nervoussystem(pp. ll4-149). Springfield, IL: Charles C
correlation, the differencedivided by the squareof the age- Thomas.
cognition correlation, and the quotient multiplied by 100 to Cerella,J.(l 985). Information processingratesi n the elderly.Psy choIog-
yield a measureof percentageattenuation of the age-related ical Bulletin. 98. 67-83.
Cerella,J.(l 99I ). Aging and information processingrate.In J.E. Birren
variance.Valuescomputed in this manner were 93Tofor the
& K. V( Schaie(Eds), Handbook of the psychologyof aging (pp.
Digit Symbol measure,85Vofor the Letter Comparison mea- 2Ol-221).SanDiego,CA: AcademicPress.
sure,987ofor the PatternComparisonmeasure,and only 46Vo Cerella,J.,Poon,L. W & Williams, D. M. 0980). Age and the com-
for the slope measure.Theseresultsthereforesuggestthat the plexity hypothesis.In L. W Poon (Ed,),Aging in the 1980s(pp. 332-
discrepanciesin the magnitudeof the attenuationof agerela- 340). Washington,DC: American PsychologicalAssociation.
tions with the slopemeasurecomparedwith other measuresare Charness,N., & Campbell,J. I. D. (1988).Acquiring skill ar mental
not simply attributable to lower measurementreliability, and calculation in adulthood: A task decomposition.Journalof Experi-
they henceserveto weakenconfidencein the interpretationof mental Psychology:General,I 17, ll5-129.
the slopeasan index ofthe magnitudeofan age-relatedreduc- Cornelius,S.W Willis, S. L., Nesselroade, J. R., & Baltes,P B. 0983).
tion in rate of cognitiveor mental processing. Convergencebetweenattention variablesand factorsof psychomet-
ric intelligencein older adults.Intelligence,7, 253-269.
Craik, E, L M. (1977).Age differencesin human memory.In J. E.
Conclusions Birren & K. W Schaie(Eds), Handbookof the psychologyof aging
(pp. 384-420). New York: Van Nostrand Reinhold.
One conclusionfrom this studyis basedon the discoverythat Ekstrom,R.8., French,J.W, Harman,H. H., &Dermen,D.(1976).Kit
the age-relatedinfluences on several measuresof memory of Factor-Referenced Cognitive Teslr Princeton, NJ: Educational
functioning were greatly attenuatedafter statisticalcontrol of TestingService.
measuresof perceptualspeed.It thereforeseemsreasonableto Hertzog,C. (1989).Influencesofcognitiveslowingon agediferences
infer that a substantialproportion ofthe adult agedifferences in intefligence.DevelopmentalPsychology,25, 636-65l.
in at leastsomemeasuresof memory appearto be mediatedby Hoyer,F W, Hoyer,W J., Treat, N. J.,& Balres,P. B. 0978-1979).
Training responsespeedin youngand elderlywomen. International
the same speedfactors that have been found to mediate age
Journalof Aging and Human Development,9,247-254.
differencesin reasoningand spatialabilities.
Kenney,D. A. 0 979). Correlationand causation.New York: Wiley.
A secondmajor conclusionis that the speedinvolvedin the Lima, S. D., Hale, S.,& Myerson,J. 0991). How generalis general
mediation of age-cognition relations does not merely reflect sfowing?Evidencefrom the lexical domain. Psychologyand Aging,
sensory and motor processesbut instead is related to how 6,4t6-425.
quickly simple mental or cognitiveoperationscan be executed. Myerson,J.,Hale,S.,WagstaflD.,Poon,L. W, & Smith,G. (1990).The
It is not yet apparent whether cognitiveoperation speedis a information-lossmodel: A mathematicaltheory of age-relatedcog-
fundamentalconstructor whetherit is a consequence of some- nitive slowing.PsychologicalReview,97, 47 5-487.
thing evenmore basic,suchasa reductionin certain attentional Pedhazur, E. J. (1982).Multiple regressionin behavioralresearch:Expla-
processes.What doesappearclear, however,is that there is a nation and prediction(2nd ed). New York: Holt, Rinehart & Win-
strong associationbetween measuresof mental or cognitive ston.
speedand the negativerelationsbetweenageand cognitiveper- Ronnberg,J. (1990).Cognitiveand communicativefunction:The ef-
fectsofchronological ageand "handicap agd' EuropeanJournal of
formance.
Cognitive Psychology,2, 253-27 3.
Salthouse,T, A. (985a). A theory of cognitive aglng Amsterdam:
North-Holland.
References
Salthouse,T. A. 0985b).Speedof behaviorand its implicationsfor
Anderson,J.C.,& Gerbing,D. V{ (1988).Structuralequationmodeling cognition. In J. E. Birren & K. Vl Schaie(Eds), Handbook of the
in practice:Reviewand recommendedtwo-stepapproach.Psychu psychologyof aging (pp. 400-426). New York: Van Nostrand Rein-
logical Bulletin, I 03, 4l l-423. hold.
Baltes,P B., Cornelius,S. W, Spiro,A., Nesselroade, J. R., & Willis, Salthouse,T, A. (l99la). Mediationof adult agedifferencesin cogni-
S. L. 0980). Integration versusdifferentiation of fluid/crystallized tion by reductions in working memory and speed of processing.
intelligencein old age.DevelopmentalPsychology, 16,625-636. PsychologicalScience,2, I 79- I 83.
Bellucci, G., & Hoyer, W J. (1975).Feedbackeffectson the perfor- Salthouse,T A. (l99lb). Theoreticalperspectives on cognitiveaging.
mance and self-reinforcingbehavior of elderly and young adult Hillsdale,NJ: Erlbaum.
women. Journalof Gerontology, 30, 456-460. Salthouse,T, A. (1992a).Mechani sms of age-cognition reIations in aduIt-
Birren,J. E. (1955).Agechangesin speedofresponsesand perception hood. Hillsdale. NJ: Erlbaum.
and their significancefor complexbehavior.ln Old agein themodern Salthouse,T A. (1992b).What do adult agedifferencesin the Digit
world (pp. 235-24'1).Edinburgh, Scorland:Livingstone. SymbolSubstitutionTestreflect?JournalofGerontology;Psychologi-
Birren, J. E. 0 965).Agechangesin speedofbehavior: Its centralnature cal Sciences.47. Pl2l-P128.
and physiologicalcorrelates.In A. T. Welford & J. E. Birren (Eds), Salthouse,T. A. (1992c).Why do adult age differencesincreasewith
Behavior,agingand the nervoussystem(pp. l9l-216). Springfield, task complexity?DevelopmentalPsychology, 28, 905-9 18.
IL: CharlesC Thomas. Salthouse,T. A. (1992d).Influenceof processingspeedon adult age
Birren, J. E.(1974).Translationsin gerontology-from lab to life: psy- differencesin working memory.Acta Psychologica,Tg,155-170.
738 TIMOTHY A. SALTHOUSE

Salthouse,T. A., & Babcock,R. L.0991). Decomposingadult age Schaie,K. W 0989). Perceptualspeedin adulthood:Cross-sectional
differences in working memory. Developmentalpsychology,27, and longitudinal studies.Psychologyand Aging, 4, 443-453.
'763-'776.
Schaie,K. W, Dutta, R., & Wiltis, S. L. (1991).Relationshipberween
Salthouse,T A., Kausler,D. H., & Saults,J. S. (l9gg). Utilization of rigidity-flexibility and cogniriveabilities in adulthood. psychology
path analytic proceduresto investigatethe role of processingre- and Aging,6,37 l-383.
sourcesin cognitive aging.Psychologyand Aging, _t,158-166. Schaie,K. W, Willis, S. L., Hertzog,C., & Schulenbere, J. E. (1987).
Salthouse,T, A., Kausler,D. H., & Saults,J.S.(1990).Age,self-assessed Effectsof cognitive training on primary ability structure. psychol-
health status,and cognition. Journal ofGerontology:psychological ogy and Aging, 2,233-242.
Sciences,4-t, Pl 56-Pl 60. Schaie,K. W, Willis, S. L., Jay,G., & Chipuer,H. (1989).Structural
Salthouse,T. A., Legg, S., Palmon,R., & Mitchell, D. R. D. (1990). invarianceofcognitiveabilitiesacrossthe adult life span:A cross-
Memory factorsin age-relateddifferencesin simple reasoning.psy- sectionalstudy.DevelopmentalPsychology,25, 652-666.
chologyand Aging,5,9-15. Steiger,J. H. (1989).EZPATH: A supplementarymodule.for SISTAT
Salthouse,T. A., & Mitchell, D. R. (l 990).Effectsof ageand naturally and SYGRAPH.Evanston.IL: SYSTAL
occurringexperienceon spatialvisualizationp erformance.Develop- Thorndike, E. L., & Lorge, I. (1944\.The teacher3wordbook oJ'30,000
mental Psychologlt,26, 845 -854. wcrrds.New York: Columbia University Press.
Salthouse,T. A., Mitchell, D. R., Skovronek,E., & Babcock,R. L. WechsleqD. (1981).TheWechslerAdult IntelligenceScale-Revised.
(1989).Effectsofadult ageand working memoryon reasoningand New York: PsychologicalCorporation.
spatialabilities. JournalofExperimentalPsychology: Learning,Mem-
or1,,and Cognitbn, 15,507-516. ReceivedJanuary8, 1992
Schaie,K. W (1985).Schaie-Tfutrstone Adult Mental Abilities Test. RevisionreceivedAprll 29,1992
PaloAlto, ( A: ConsultingPsychologists Press. AcceptedOctober13,1992t